lnk_pipeline_crossings: missing PSCIS<->modelled 100m-instream auto-snap layer

[NewGraphEnvironment]

Problem

lnk_pipeline_crossings v0.32.0 carries the PSCIS<->modelled crossing linkage only from the xref CSV (pscis_modelledcrossings_streams_xref). It's missing bcfp's automatic 100m-instream snap layer that produces the bulk of PSCIS<->modelled linkages.

See research/bcfp_table_map.md for the full mechanism analysis. Short version: bcfp's 02_pscis_streams_150m.sql (at smnorris/bcfishpass@v0.7.14-125-g6e9cf1c, current tunnel state bcfishpass.log.model_run_id=121 rebuilt 2026-05-05) auto-matches each PSCIS to its nearest modelled crossing within 100m instream distance on the same stream. The xref CSV layers manual overrides on top. link skips the auto-snap entirely.

Evidence (Phase A bcfp parity, 2026-05-10)

WSG	extra <schema>.barriers_anthropogenic vs tunnel	mapping_code parity (worst species)
ADMS	+89	bt 98.53% (211 diffs)
BULK	+1391	bt 76.30% (9357 diffs)
WILL	+N (unmeasured)	bt 85.48% (2490 diffs)
PARS	cross-WSG dnstr (#152)	bt 56.16% (16445 diffs)

The +1391 in BULK is the largest single contributor to BULK's mapping_code drift. Closing this gap is expected to bring all WSGs' mapping_code parity into >=99% (modulo cross-WSG dnstr, handled by #152).

Dependencies

Blocked by fresh#206 -- the generic frs_point_match primitive (point-to-point match on FWA network within instream distance). Once that ships, this issue is a thin integration.

Proposed solution

Once fresh ships frs_point_match:

1. Add <schema>.pscis build step in lnk_pipeline_crossings, mirroring bcfp's bcfishpass.pscis:
   • Source: <schema>.pscis_assessment_snapped (already produced by lnk_points_snap)
   • Call fresh::frs_point_match with table_a = <schema>.pscis_assessment_snapped, table_b = fresh.modelled_stream_crossings, table_to = <schema>.pscis, distance_max = 100, table_a_id_col = "stream_crossing_id", table_b_id_col = "modelled_crossing_id"
   • Result: PSCIS rows with modelled_crossing_id populated where auto-snap found a match
   • Apply xref-CSV overrides on top (xref overrides auto-snap matches)
2. .lnk_crossings_union reads from <schema>.pscis for the PSCIS branch (instead of <schema>.pscis_assessment_snapped).
3. The modelled-branch exclusion changes: drop the xref-based WHERE NOT IN (xref) and replace with WHERE NOT IN (SELECT modelled_crossing_id FROM <schema>.pscis WHERE modelled_crossing_id IS NOT NULL) -- picks up both auto-snap and xref-derived linkages from the same source.

Acceptance

• <schema>.pscis.modelled_crossing_id populated where bcfp's bcfishpass.pscis.modelled_crossing_id is, for ADMS / BULK / WILL / PARS (within tolerance)
• <schema>.crossings modelled-source row count matches bcfishpass.crossings for ADMS / BULK / WILL / PARS (within tolerance)
• compare_bcfp_mapping_code.R --wsgs=ADMS,BULK,WILL,PARS mapping_code parity >=99% for all species except PARS BT (cross-WSG, separately handled by Unified <persist_schema>.barriers (province-wide) with blocks_species predicate #152)
• Roxygen + lintr clean

Out of scope

• Cross-WSG dam_dnstr_ind -- Unified <persist_schema>.barriers (province-wide) with blocks_species predicate #152
• lnk_pipeline_species / lnk_presence alignment -- lnk_pipeline_species vs lnk_presence: align presence definitions for mapping_code parity #153
• Extending PSCIS to all 4 BCDC views (currently only pscis_assessment_svw -- bcfp uses all 4) -- separate question