feat: WeightedRanker with IDF tag weighting (Jaccard fallback, train/save/load)

[laurentketterle-hub]

Summary

Adds WeightedRanker — a trainable tag-based plant species ranker that extends the existing ToyPlantIdentifier (Jaccard baseline) with IDF-like weighted tag scoring.

What's new

• src/plantguide/models/weighted_ranker.py — WeightedRanker class with:

  • train() — compute per-tag IDF weights from the species catalog (sklearn TF-IDF when available, pure-Python IDF fallback otherwise)
  • save(path) / load(path) — persist/restore learned weights as JSON
  • identify(tags, top_k) — 70/30 blend of weighted score + Jaccard for stable ranking; pure Jaccard fallback when weights aren't trained
  • Compatible interface with existing ToyPlantIdentifier

• src/plantguide/models/__init__.py — exports WeightedRanker

• tests/test_weighted_ranker.py — 10 pytest tests covering:

  • Pure-Python IDF training and idempotency
  • Save/load round-trip + partial-tag loading
  • Weighted identification (correct top match, score fields)
  • Jaccard fallback (identical results to ToyPlantIdentifier when untrained)
  • Edge cases (empty query, single tag)
  • Integration: top-1 hit rate on labeled samples vs Jaccard baseline

Results

59 passed in 96s (all existing + 10 new, zero regressions)

Fixes #11

[laurentketterle-hub]

/attempt 11

[laurentketterle-hub]

Friendly ping 👋 — this PR has been open for 3 days. Is there anything needed for review? Thanks!