…y.rs #423) (#4184)

* feat(viz): KPI overview row, icicle charts & layout uirevision (plotly.rs #423)

Adopt dathere/plotly branch `feature/trace-type-additions` (PR plotly/plotly.rs#423,
a strict superset of the prior pin) and use the new API in `viz`:

- viz smart now leads with a full-width KPI overview row of plotly `Indicator`
  tiles: a built-in completeness gauge plus the headline numeric measures (sum for
  extensive, mean for intensive). A measure becomes a gauge when the data
  dictionary supplies a validated `x-qsv.gauge_range` (drawn only when the range
  actually contains the observed value; a range also forces the mean so it lands
  in-scale) and a "vs target" delta when it supplies `x-qsv.target` (a
  semantically-justified goal, never a fabricated prior-period baseline). Renders
  in both the inline and typed-grid paths and stays invisible to --max-charts,
  the panel-count notices, and the render-path thresholds (it is a domain band,
  not a cartesian subplot). HTML-only.

- Add `viz icicle` subcommand and `--hierarchy-style icicle`, a level-aligned
  sibling of treemap/sunburst.

- Set layout-level `uirevision` centrally in apply_theme so user zoom/pan/
  selection/legend state survives the Plotly.newPlot re-renders that inject the
  fullscreen modebar button and recolor on theme toggle.

describegpt emission of gauge_range/target is deferred; until then the hints are
consumed from any dictionary that carries them, with completeness + number tiles
working with no dictionary.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(viz): don't orphan the 8th cartesian axis behind the KPI row (roborev #3602)

The leading KPI row is domain-positioned and consumes no cartesian axis, but it
still occupies panel index 0 in `smart_grid_parts`, so `pos = n + 1` labelled the
cartesian panels behind it x2..x9. With exactly MAX_SUBPLOTS (8) cartesian panels
on the typed-grid path the 8th chart landed on x9, which the typed `Layout`
(x1..x8) has no slot for — `assign_typed_axis` silently drops it, orphaning the
trace onto the default axis.

Subtract a one-slot `axis_offset` when the leading panel is the KPI row so the
cartesian axes number x1..x8. The KPI row and geo panels never coexist in this
assembler (geo forces the inline path for HTML; image exports carry no KPI row),
so a single leading offset is sufficient. Adds a regression test that drives the
exact 8-cartesian-panel boundary via `--max-charts 8`.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(viz): regenerate gallery with KPI overview row & add icicle showcase

The smart dashboards now lead with a KPI overview row (completeness gauge +
top-measure number tiles); regenerated the 7 affected smart_*.html figures so
they reflect it. Added a dedicated `viz icicle` example to gen_gallery.py so the
new chart type is showcased (gallery is now 38 figures).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(viz): correct icicle default-orientation comment; document pre-gen KPI-row lag (roborev #3604)

The icicle code comment claimed a top-down default, but plotly leaves
tiling.orientation at horizontal ("h") — root on the left, children fanning
right (confirmed in plotly.rs Tiling docs + live render). The gallery
description was already accurate; fixed the misleading code comment instead.

Also documented, per developer deferral, that the LLM-dependent pre-generated
dashboards intentionally lag the KPI/Completeness row until describegpt emits
the gauge_range/target dictionary hints and they can be regenerated in one pass.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>