Add Confidence field and enrich deck descriptions with selection stats

[SKOHscripts]

Summary

Two follow-up improvements on top of the sampling refactor:

1. New Confidence Anki field

• The Bayesian quality score computed during selection (_quality, combining Popularity + NbPlays, confidence-weighted) is now written to each puzzle CSV as a Confidence column.
• Added a Confidence field to the Anki note model (appended last to keep field order stable for existing collections).
• Displayed on the card back as a ★ {score} pill so users can gauge how community-validated a puzzle is.

2. Richer deck descriptions

Each generated sub-deck description now mirrors the generation-time coverage report:

• Summary line: puzzle count · ELO range/average · average popularity
• Woodpecker mastery estimate: time for a first full pass at ~20 new puzzles/day (from the README), then accelerated repetition cycles — a motivational one-liner per deck
• Coverage block: unique themes/motifs covered, distinct themes/motifs in tranche, motif coverage % (+ all-theme coverage %)
• Full theme-frequency breakdown (every theme, sorted by frequency)

Example:

📊 1200 puzzles · Rating 800–999 (avg 900) · Popularity avg 92%

🔨 Woodpecker plan: at ~20 new puzzles/day, a first full pass takes about 60 days (~9 weeks). ...

📈 Coverage:
- Unique themes covered: 69 (motifs: 54)
- Distinct themes in tranche: 72 (motifs: 54)
- Motif coverage: 100.0%  (all-theme coverage: 95.8%)

🎯 14 themes by frequency:
  • endgame: 274 puzzles
  ...

Changes

• lichess_optimized_puzzles_datasets.py: write Confidence column; _format_confidence helper.
• build_apkg.py: new field + _row_to_note mapping + SAMPLE_CARDS; rewritten _build_description (now takes a stats dict) with _mastery_sentence and _coverage_lines helpers.
• templates/back.html: confidence pill.
• tests/build_apkg_test.py: updated to the new description format and stats param; added breakdown/mastery tests.
• README.md: documents the field and the enriched descriptions.

Verification

• pytest tests/ — 60 passed
• pylint 10.00/10 (CI flags), mypy clean

---

Generated by Claude Code