Confirm link-built crossings match bcfp's complete crossings table (parity)

[NewGraphEnvironment]

link builds crossings from primitives itself — lnk_pipeline_crossings → .lnk_crossings_union (PSCIS snap/score/xref + fresh.modelled_stream_crossings + CABD), re-implementing bcfp's crossings assembly (02_pscis_streams_150m.sql + 04_pscis.sql). Because we reproduce that logic instead of consuming bcfp's output, we should confirm link's built <schema>.crossings is equivalent to bcfp's complete crossings table (bcfishpass.crossings / crossings_vw).

bcfp's complete crossings table carries a lot of derived info and is built by many scripts — it's the natural parity reference for validating our primitives-build, now and as bcfp evolves. (The newgraph crossings.csv we dumped is NOT a model input — the models build from primitives — but this complete table is our parity yardstick.)

Scope

• Compare link <schema>.crossings vs bcfishpass.crossings_vw on a sample of WSGs: row coverage, per-branch counts (PSCIS / modelled / CABD), barrier_status, network positions (blue_line_key + downstream_route_measure). Quantify divergence.
• Make it a repeatable parity check, alongside the existing mapping_code / accessible_km cross-sections.

Related / context

• Supersedes the misdirected Consume weekly crossings.csv; repoint pipeline off fresh #231 (which assumed the pipeline consumes fresh's crossings.csv — it doesn't). Lessons + full trace archived at planning/archive/2026-07-issue-231-crossings-from-primitives/.
• Freshness: primitives are loaded manually by data-raw/snapshot_bcfp.sh (last modelled-crossings load ~2026-05-26). Automating that refresh weekly is a related follow-up (may split into its own issue).
• db_newgraph#16 tracks reconsidering the now-model-unneeded crossings.csv dump.