frs_habitat_classify: recycle accessibility by threshold group

[NewGraphEnvironment]

Problem

frs_habitat_classify runs the NOT EXISTS accessibility query once per species. Species sharing the same access threshold (CO/CH/SK all at 15%) run identical queries — 3x redundant work. For BULK (32k segments, 7 species), classify takes 5+ min per species.

Two optimizations needed

1. Recycle accessibility by threshold group

Compute accessibility ONCE per unique threshold, store result, map to species:

Threshold 15% → accessible_15 (computed once)
  CO, CH, SK, PK → use accessible_15
Threshold 20% → accessible_20  
  ST → use accessible_20
Threshold 25% → accessible_25
  BT, RB → use accessible_25

3 queries instead of 7. Same results.

2. Index breaks table on (label, blue_line_key)

The NOT EXISTS queries filter by b.label IN (...) then join on b.blue_line_key. A composite index would make this instant. Currently .frs_index_working indexes label and blue_line_key separately.

Impact

Current ADMS (5 species): ~183s
Estimated with recycling: ~90s (2 threshold queries instead of 5)
Estimated with recycling + composite index: ~30s

Relates to

• Unified network segmentation: frs_network_segment + frs_habitat_classify #83 — unified pipeline
• Profile and optimize Phase 2 species classification performance #72 — performance optimization