feat: Phase 1 Quick Wins - product type modifiers and LLM edge cases

Implements Phase 1 of classifier redesign:

Query Parser (query-parser.service.ts):
- Add PRODUCT_TYPE_MODIFIERS constant (35 modifiers)
- Processing state: instant, soluble, roasted, etc.
- Form indicators: powder, beans, leaves, frozen, etc.
- Coffee/tea varieties: arabica, robusta, green tea, etc.
- Function indicators: for vehicles, automotive, etc.
- Update parseQuery() to extract productTypeModifiers

LLM Navigator (llm-navigator.service.ts):
- Add shouldExcludeByModifier() for strict filtering
- Coffee exclusion: instant/extract vs roasted/raw
- Variety exclusion: Arabica vs Robusta
- Update filterOptionsByKeywordsSimple() with modifier filtering
- Add critical edge cases to LLM prompt:
  - Coffee classification (Ch.09 vs Ch.21)
  - Function over material rules
  - Variety and grade handling
  - "Other" category rules

Data files added:
- chapter-triggers.json (chapter prediction rules)
- elimination-rules.json (eliminative logic rules)

Cleanup:
- Remove obsolete write-chapter-predictor.js script

Test results:
- ✅ "instant coffee" correctly routes to Ch.21
- ⚠️ "brake pads" needs Phase 2 work for function-over-material

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: AryanBV