viz smart: add a Quarter time bucket + cadence-aware bucket selection (describegpt-inferred temporal cadence)

[jqnatividad]

Context

Follow-up from PR #4215 (viz smart panel-heights + demo-dataset polish).

viz smart's time bucketing (src/cmd/viz.rs, enum TsBucket { Day, Week, Month } ~L13132) has three granularities and picks between them purely by the date column's observed span (ts_bucket_for_span, from the stats-cache min/max). There is no Quarter (or Year) bucket, and the selection ignores the data's native cadence.

This surfaced with the quarterly regions_growth.csv demo: a quarterly date column spanning ~15 months bucketed to weekly/monthly, so the "Records by <period>" time-series showed data every 3 months with empty periods between — it read as sparse/"weird". In PR #4215 the pragmatic fix was to make that demo monthly and dense instead, which sidestepped (but didn't solve) the underlying gap.

Important nuance discovered in #4215

The TimeSeries panel renders on a date x-axis (is_date → AxisType::Date, viz.rs:styled_x_axis). So a Quarter bucket by itself would not change the visual — quarterly points would still sit 3 months apart on a continuous date axis. To actually help, a Quarter bucket must be paired with either:

• quarter-aware tick labels (YYYY-Qn) and/or a categorical x-axis (one evenly-spaced slot per bucket), and
• cadence-aware bucket selection so quarterly data is bucketed by quarter in the first place (span-based selection alone picks Week/Month for a ~1–2 year span).

Proposal

1. Add TsBucket::Quarter (and consider Year) with matching arms in:

   • ts_bucket_key (truncate to quarter-start month: 1/4/7/10, day 1)
   • ts_bucket_label (YYYY-Qn)
   • ts_bucket_word ("quarter")
   • choose_anim_bucket (~L13535) so the T3 Gapminder animation can use quarter frames
   • ts_bucket_for_span thresholds (Quarter/Year for multi-year spans)

2. Cadence-aware selection. Rather than span alone, pick the finest bucket that doesn't leave most periods empty — e.g. cap fineness by the date column's distinct-value count / typical spacing (the stats cache already has cardinality; choose_anim_bucket already does a distinct-count-based pick for the animation). Quarter-aligned dates → quarter buckets.

3. Rendering. Give bucketed TimeSeries/count panels quarter-appropriate ticks (or a categorical axis) so evenly-spaced quarters replace the gappy date-axis look.

The describegpt angle ("an additional temporal concept")

The concept vocabulary (src/cmd/describegpt/dictionary.rs, CONCEPT_VOCAB) already has time.* role concepts (time.date, time.event_timestamp, time.created_at, …) but nothing describing a date column's cadence/granularity (daily / weekly / monthly / quarterly / yearly). describegpt already infers dataset-level grain and temporal_coverage — inferring a date column's native cadence fits the same machinery.

Idea: describegpt infers a temporal-cadence attribute for date columns (either a new column attribute alongside concept/role, or a time.* cadence hint), and viz smart consumes it via the dictionary (like it already consumes measure.*/geo.* concepts) to drive bucket selection deterministically — instead of re-deriving cadence from the raw data each run. This would also let --dictionary/--smarter runs bucket correctly even when the span heuristic would guess wrong.

Open question: model cadence as a column attribute (cleaner — cadence is orthogonal to the timestamp's role) vs. a time.* concept token. Leaning attribute, but worth deciding during design.

Acceptance criteria

• A quarterly dataset (e.g. a quarterly regions_growth-style CSV) produces a viz smart time-series bucketed and labeled by quarter, with no empty inter-quarter gaps.
• choose_anim_bucket can select quarter frames for the T3 animated bubble.
• (Stretch) describegpt infers a date column's native cadence; viz smart honors it from the dictionary.
• Tests: ts_bucket_key/label/word for Quarter; cadence-aware selection on quarterly vs daily data; a gallery demo (restore a quarterly regions_growth variant) verifying the gap-free quarter axis.

Critical files

• src/cmd/viz.rs — TsBucket, ts_bucket_for_span/_key/_label/_word (~L13132–13176), choose_anim_bucket (~L13535), TimeSeries bucket selection (~L13373), styled_x_axis date-axis handling.
• src/cmd/describegpt/dictionary.rs — CONCEPT_VOCAB / temporal concepts; src/cmd/describegpt.rs — grain/temporal_coverage inference.