Add falls to break_order — fix segmentation drift (Fixes #96)

[NewGraphEnvironment]

Summary

falls was loaded into <schema>.falls and consumed for access gating + obs/habitat lift but never used as a segmentation break source, despite R/lnk_pipeline_break.R lines 10-13 documenting bcfp's order as observations → gradient_minimal → falls → barriers_definite → habitat_endpoints → crossings. When two falls sat close together with no other break source between them, the resulting fresh.streams segment spanned the second fall, leaving its upper portion incorrectly classified as accessible.

Fix

• falls = paste0(schema, ".falls") added to source_tables in R/lnk_pipeline_break.R
• "falls" added to break_order default vector (between gradient_minimal and barriers_definite)
• "falls" added to pipeline.break_order in both inst/extdata/configs/{bcfishpass,default}/config.yaml
• Falls are NOT minimal-reduced — each fall is its own barrier (unlike gradient barriers which go through frs_barriers_minimal)

Verification

HORS BLK 356357296 evidence case: segment 12671 (1447 m straddling fall #2 at DRM 67565) split into 12677 (17 m below, accessible) + 12678 (1429 m above, accessible=FALSE). Log: data-raw/logs/20260501_27_preflight_hors_falls_break.txt.

HARR BLK 356361157 (7 falls in 13 km): all 7 fall positions now have segment breaks. Log: data-raw/logs/20260501_28_preflight_harr_falls_break.txt.

4-WSG regression (HARR/HORS/LFRA/BABL) vs pre-#96 baseline at data-raw/logs/provincial_parity/:

WSG	diff_rows	Total km removed	Notes
HARR	2	−1.27	BT only
HORS	2	−2.89	BT only; affected segment is edge_type 1250 (Horsefly River construction line) so rearing_stream metric unchanged
LFRA	14	~−6	7 species × ~0.43 km each
BABL	12	~−12	4 species × 0.94–1.59 km each

All deltas negative — fix correctly removes segments above falls. Direction is structurally right regardless of whether each removal closes or slightly deepens the bcfp gap.

Test plan

• 2 new unit tests in tests/testthat/test-lnk_pipeline_break.R (33 PASS, was 29)
• HORS verification on the issue's evidence case
• HARR sanity check on a 7-fall BLK
• 4-WSG regression
• research/bcfishpass_comparison.md § "falls in break_order (Falls not used as segmentation break source — implementation drift from documented break_order #96)" with the evidence trace + regression table
• DESCRIPTION + NEWS bump 0.22.0 → 0.23.0

Map cache helper data-raw/maps/_lnk_map_compare.R hardened alongside — stale 0-row caches (left when the pipeline runs for one WSG and the map is rendered for another) now refetch instead of erroring on missing CRS.

🤖 Generated with Claude Code