v0.23.0 — Per-tile quality (uncertainty) API + v3.1.1 catalog

Added — per-tile quality (uncertainty) API

• stats.add_uncertainty_columns(df, *, min_votes=5, circular_ratio_limit=0.8)
  enriches a fan or blotch catalog (kind inferred from its columns) with four
  per-marking columns derived from the cluster standard deviations: pos_std
  (hypot(x_std, y_std)), size_cv (a mean relative std — distance/spread
  for fans, radius_1/radius_2 for blotches), angle_usable (False for
  near-circular blotches, where orientation is ill-defined) and scatter_ok
  (the n_votes >= min_votes gate).
• stats.tile_quality(kind="both", version=None, *, min_votes=5, agg="median", ranks=True) aggregates to one row per tile_id with the support and
  scatter column groups kept strictly separate: scatter (angle_scatter,
  pos_scatter, size_scatter) is computed only from markings that pass the
  vote gate, so weak support can never masquerade as disagreement, and vote
  counts never divide the scatter numbers (no SEM). Optional cap-wide
  percentile support_rank/scatter_rank make the incommensurate units
  combinable.
• stats.classify_tile_quality(tq, *, support_thresh=50, scatter_thresh=50)
  labels tiles consistent / contested / sparse / noisy from the two rank
  axes; tiles with no gated marking (hence no scatter estimate) are left <NA>
  rather than guessed.
• New tutorial 03a_tile_quality introduces the API with live v3.1
  examples; tests in tests/test_stats.ipynb.

Added — coverage & plotting helpers

• coverage.pct_summary(coverage, pcts=...) returns a percent-scaled
  percentile+mean summary; coverage.attach_tile_metadata(per_tile, *, version="v3.1") adds roi_name/Ls/MY per obsid.
• plotting.plot_random_tiles_with_min_fans and
  plot_random_tiles_with_min_blotches sample and plot random tiles meeting
  a minimum marking count.

Changed — v3.1 catalog re-published with per-marking cluster std columns

• The v3.1 fan/blotch catalogs are re-published as Zenodo v3.1.1
  (doi:10.5281/zenodo.21917907), adding the seven per-marking cluster standard
  deviations (x_std, y_std, angle_std, plus distance_std/spread_std
  for fans and radius1_std/radius2_std for blotches) that feed the new
  uncertainty API. The change is strictly additive — every pre-existing
  column and value is byte-identical — so get_fan_catalog("v3.1") /
  get_blotch_catalog("v3.1") transparently gain the columns with no break to
  existing code. The io v3.1 registry is repointed and all v3.1 files unified
  onto the single new DOI.
• ReleaseManager.FAN_COLUMNS_AS_PUBLISHED / BLOTCH_COLUMNS_AS_PUBLISHED
  now retain the std columns, so future catalog production carries them through
  merge_all rather than dropping them at publish.