Break streams only where a present species' access changes (both configs); bcfishpass config takes the order-1 exception #326

If done: both configs break streams only where a present species' access changes, which is how bcfishpass breaks them. Segment counts roughly halve, and BT habitat stops flipping on breaks that gate no fish present. The bcfishpass config also takes bcfishpass's order-1 rearing exception. If never: BT-only groups (no 15 / 20 % species present: PCEA, FINA, PARS) keep 71–85 km of two-way BT rearing disagreement under a 0.5–1.9 % net (#324), and every group carries 2–3× the segments it needs.

Step-by-step comparison and history: research/bcfp_segmentation_parity.md.

Why the breaks differ

• bcfishpass (smnorris/bcfishpass@620dbe4, model/01_access/01_model_access_natural.sh) builds each model's barriers after its observation and habitat lifts, keeps the downstream-most per flow path, then breaks there. Models build only where their species are present.
• link (.lnk_pipeline_prep_minimal()) builds a set for every configured species whatever is present (Gradient classes: derive from parameters_fresh, optional override arg #45), and since BT/ST accessible_km over-credit: access segmentation drops the accessibility frontier #223 breaks at every position. Before BT/ST accessible_km over-credit: access segmentation drops the accessibility frontier #223 it reduced, but before the lifts, so a lifted mainstem barrier pruned real tributary frontiers. BT/ST accessible_km over-credit: access segmentation drops the accessibility frontier #223 fixed that by dropping the reduction instead of moving it after the lifts.
• PCEA: link 106,079 segments vs bcfishpass 37,957; link holds all but 1 of bcfishpass's 24,393 breaks plus 68,118 extra.

Scope

1. Break where access changes, both configs. Per species present in the WSG, reduce its post-lift barrier set (barriers_<sp>_access) to the downstream-most position per flow path, and break there. No config switch: it is the general rule, and default follows.
2. Order-1 rearing exception in the bcfishpass config. bcfishpass admits an order-1 line with parent order ≥ 5 whose known width is below the minimum (BT, CH, CO, ST, WCT; load_habitat_linear_bt.sql:90-97); NULL widths never qualify. Needs a width test in fresh's frs_order_child(); then set rear_stream_order_bypass in the bundle. PARS BT: −24.78 km. default's position is a separate issue.
3. Optional, lowest priority: strict > on the spawning size minimum in the bcfishpass config (bcfishpass channel_width > spawn_channel_width_min and mad_m3s > spawn_mad_min; fresh tests >=). default keeps >= (research/default_vs_bcfishpass.md §6). BT 1.8–5.3 km per group, all at exactly 2.0 m. Needs a fresh option; drop it if it costs more than that.

Not in scope (known divergences, annotated)

• ct_dv_rb: bcfishpass builds it only in a fixed list of 14 groups (CCIRA, LDEN / UDEN, COWN), outside the groups we model. link has no such model.
• The 1.8 km PCEA access residual (three segments at subsurface-flow or falls barriers).

Before landing

• Confirm on FINA blk 359209845 that reducing the post-lift set keeps the frontier at 3,835 m (BT/ST accessible_km over-credit: access segmentation drops the accessibility frontier #223's case).
• Re-verify accessible-km parity, 11 WSGs × 8 species (data-raw/parity_crosssection.R).
• Measure BT habitat against bcfishpass with data-raw/habitat_diff_segments.R on PCEA / FINA / PARS / LKEL: segmentation churn and the width-exception km should go.
• State the run decisions before re-modelling: every persisted group re-segments, plus the post-consolidate recompute.
• Update the stale "matches bcfishpass" comment in R/lnk_pipeline_prepare.R, close Rename gradient_barriers_minimal -> gradient_barriers_break (no longer minimal-reduced) #225 (the table is minimal again), and retire taxonomy hors-class-stream-order-bypass.

Relates to #324, #223, #45, #225.