ggpointless 0.3.0

New features

• New geom_unit_bar(), geom_unit_col(), and geom_unit_histogram()
  draw isotype / pictogram bar charts, where each bar is a stack of discrete
  unit cells (one cell = one observation by default). geom_unit_bar() counts
  observations like geom_bar(); geom_unit_col() uses pre-computed y like
  geom_col(); geom_unit_histogram() bins a continuous variable like
  geom_histogram(). Fractional y values produce a partial cell at the
  outer edge. All three work with coord_equal(), coord_polar(),
  coord_radial(), coord_flip(), and faceting. #15

• New label_cells(cell_size) helper pairs with cell_size to relabel
  the value axis in cell counts instead of raw units: pass it to
  scale_y_continuous(labels = ...) so a tick at y = 200 displays as 2
  when cell_size = 100.

• New geom_gridline() draws (panel)grid lines as a layer on top of other
  geoms. Reads positions directly from the trained scales (no manual breaks
  needed), inherits line styling from
  theme(panel.grid.major.*) / panel.grid.minor.* for any property not
  explicitly overridden, and supports coord_polar() / coord_radial(). Use
  grids = "x" / "y" / c("x", "y") to pick axes; minor = TRUE to draw
  minor lines too. To remove the underlying theme grid, compose
  with + theme(panel.grid = element_blank()). #12

• New geom_path_fade(), geom_line_fade(), and geom_step_fade() draw
  paths, lines, and step functions with a linear alpha gradient along their
  length, so one or both ends fade to transparent. The fade_direction argument
  controls which end(s) fade ("end", "start", or c("start", "end"));
  alpha_fade_to sets the target alpha.

• New geom_segment_fade() draws individual line segments like
  geom_segment() but fades each segment along its own direction — the
  gradient follows the segment from (x, y) to (xend, yend), so it works at
  any angle. Accepts the same fade_direction and alpha_fade_to arguments as
  geom_path_fade(). #11

• New geom_curve_fade() draws Bézier curves like geom_curve() but with
  an alpha gradient along the curve direction. The fade follows from start to
  end point, so curves fade at any angle. Uses the same fade_direction and
  alpha_fade_to arguments as geom_segment_fade(), with Porter-Duff
  compositing for smooth gradients. Falls back to semi-transparent curves on
  unsupported devices. #11

• New geom_abline_fade(), geom_hline_fade(), and geom_vline_fade()
  draw reference lines (diagonal, horizontal, vertical) with an alpha gradient
  along the line direction. They mirror the ggplot2 annotation pattern: pass
  slope/intercept, yintercept, or xintercept directly for constant
  lines, or supply data and mapping for facet-varying lines. Under
  non-linear coordinate systems (coord_polar(), coord_radial()) the fade
  follows the curve that the coord transform produces — geom_hline_fade()
  fades around a circle, geom_vline_fade() fades along a ray, and
  geom_abline_fade() fades along the resulting arc.

• New geom_rect_fade() draws arbitrary rectangles with a
  linear alpha gradient that fades one edge to transparent. The fade_direction
  argument controls the gradient direction: "vertical" (default) fades from
  opaque at the top to transparent at the bottom; "horizontal" fades from
  opaque at the left to transparent at the right. Supports a radius argument
  for rounded corners too. #13

• New geom_col_fade(), geom_bar_fade(), and geom_histogram_fade()
  draw bar charts with a vertical alpha gradient that fades from opaque at
  the peak to transparent at the baseline. Additionally these geoms support a
  radius to draw bar charts with rounded corners
  The rendering tier is chosen at draw time: gradient fill on capable devices, flat
  semi-transparent fill on pdf()/postscript(). #14

• New geom_freqpoly_fade() draws a filled frequency polygon — the area under
  the geom_freqpoly() line — with the same fading gradient as
  geom_area_fade(). Paired with stat_bin(), so all binning parameters
  (bins, binwidth, center, boundary, …) are forwarded.

• New geom_density_fade() draws a kernel density estimate with a fading
  gradient, using geom_area_fade() paired with stat_density(). Accepts all
  smoothing parameters (bw, adjust, kernel, bounds, …).

• New geom_ridgeline_fade(), geom_ridgeline_density_fade(),
  geom_ridgeline_histogram_fade(), and geom_ridgeline_freqpoly_fade() draw
  ridgeline plots — overlapping ridge shapes at different vertical offsets — with
  a vertical alpha gradient that fades from opaque at each ridge's peak to
  transparent at the baseline. The alpha_scope argument controls how alpha is
  scaled across ridges. geom_ridgeline_fade() takes explicit (x, y, height);
  the three other forms compute height from a stat (smooth density, stepped
  histogram, polyline frequency polygon) so users only supply (x, y). All four
  handle negative heights (dips below the baseline) with a bidirectional gradient.
  The scale argument defaults to NULL, which auto-scales the layer so
  the tallest ridge overlaps its neighbour by ~50% (the canonical "mountain
  range" ridgeline look). The auto-resolved value is reported via a
  cli::cli_inform() so you have a starting point if you want to override. #19

• New PositionRidgeline (callable via position = "ridgeline") is the
  position adjustment that converts a (y, height) mapping into the
  ribbon-style (ymin, ymax) form GeomRibbon expects, so each ridge sits
  on its own y-baseline. Used as the default position by
  geom_ridgeline_fade(), but also composes with vanilla
  [ggplot2::geom_ribbon()] for users who want the layout without the fade.

Breaking changes

• geom_point_glow(glow_size = X) now interprets X at face value in
  ggplot2 size units, matching the size aesthetic of
  [ggplot2::geom_point()]. Previously X was wrongfully multiplied by 3
  before rendering, so existing user-supplied values render roughly
  three times smaller than in 0.2.0. The default (glow_size = NA) is
  unchanged and still renders at nine times the point's size.

• Removed geom_lexis(point_size = ...), deprecated since 0.1.0 in favour
  of size. Code still passing point_size will now error with
  unused argument; rename the argument to size.

Bug fixes and improvements

• stat_chaikin() no longer renders nothing without a message when the
  smoothed data has missing values in different rows of (x, y) than in a
  numeric extra aesthetic (e.g. aes(fill = z) where z has its own NAs).
  Previously each numeric column was de-NA'd independently, the resulting
  per-column outputs had different lengths, the assignment errored, and
  ggplot2's stat machinery silently swallowed the error — leaving an empty
  layer. The stat now pre-filters complete cases across all numeric columns
  once and warns about dropped rows in the standard ggplot2 way (suppress
  via na.rm = TRUE). #18

• stat_fourier() / stat_catenary() / stat_arch() no longer crash
  ggplot2's limit expansion when paired with coord_transform(y = "log10")
  (and other restricted-domain transforms such as "log", "sqrt").
  The stats now drop those rows with a single helpful warning that points users
  at scale_y_log10(), which transforms before the stat runs and avoids the
  issue entirely.

• stat_pointless() / geom_pointless(): when a single observation matches
  multiple location criteria — e.g. the last point is also the maximum —
  after_stat(location) now carries a composite label (e.g. "last, maximum")
  instead of silently dropping the secondary labels. Previously only the first
  matching label in iteration order was kept. Row order still follows the order
  given in location (canonical "first", "last", "minimum", "maximum"
  for "all").

• geom_area_fade(): now renders alpha gradient when + coord_flip() is
  applied. Previously the geom did not render any fill without a message or
  warning shown to users. #17

• geom_area_fade(): on devices without gradient support (base pdf(),
  postscript()) the informational message now fires for solid-fill plots too
  — users were previously silent about the lost vertical fade unless fill
  was mapped to a variable. Wording is tailored to what was actually lost
  (colour gradient vs. vertical fade). Devices that support gradients but not
  compositing still stay silent for solid-fill plots (tier 2 renders the
  vertical fade faithfully).

• geom_area_fade(): the global_max_abs scan that drives alpha_scope = "global" now handles Date and POSIXct value axes. Previously both
  failed is.numeric() and silently fell back to global_max = 1; plots
  built through the normal scale pipeline were unaffected (values are
  already numeric by draw time), but direct draw_panel() calls with raw
  Date / POSIXct ymax are now robust.

• geom_area_fade(): the legend key (.draw_key_area_fade()) now validates
  alpha_fade_to via the shared .check_alpha_fade_to() helper, matching
  the validation used by setup_params(). A guide that constructs a key
  outside the normal setup pipeline now aborts on out-of-range values
  instead of silently producing an invalid gradient.

• geom_area_fade(): fixed alpha overflow for stacked areas with
  alpha_scope = "global" (the default). The reference max was computed
  from the pre-stacking y values, so the top ribbon's alpha exceeded 1
  and was silently clamped to fully opaque, defeating the fade. The
  reference is now taken from post-position-adjustment data, so equal
  rendered |y| maps to equal opacity as documented. No effect on
  position = "identity" or alpha_scope = "group".

• geom_area_fade(): now accepts both integer and floating-point alpha_fade_to
  values (e.g. 0L, 0.5). Previously integer input was rejected.

• geom_area_fade(): fixed has_outline check to handle coord_polar() without
  crashing when outline colour is a vector.

• geom_area_fade(): fixed duplicate comp_stops that could arise when values
  were clipped to [0, 1] bounds (e.g. when val_hi is far outside the panel).
  Identical stops are now de-duplicated before gradient construction.

• geom_area_fade() under coord_flip() now rotates its alpha gradient to
  follow the rendered visual axis, instead of staying vertical and visually
  disappearing against horizontal bars. The gradient is now anchored to the
  rendered geometry via flipped_visual = xor(flipped_aes, CoordFlip), so
  coord_flip() and orientation = "y" produce equivalent output.

• New custom legend key glyphs for geom_fourier() (sine wave),
  geom_catenary() (hanging curve), and geom_arch() (arch curve) provide
  custom visual cues in the legend.

• All draw_key_*() functions now consistently use ggplot2::gg_par() instead
  of grid::gpar() for proper theme resolution in ggplot2 v4.0+.