Phase 2: enrich streams with mad_m3s and add mad predicate to rule evaluator

[NewGraphEnvironment]

Problem

Habitat eligibility rules (#113 Phase 1) handle edge_type, waterbody_type, gradient, and channel_width predicates. They cannot evaluate mad (mean annual discharge) predicates because the segmented streams table produced by frs_network_segment() (fresh 0.11.0) does not carry a mad_m3s column.

bcfishpass v0.5.0 uses MAD-based habitat classification for many southern BC watershed groups (BULK, LDEN, etc.) — the parameters_habitat_method table sets model = 'mad' for those groups. Per-species MAD thresholds in parameters_habitat_thresholds.csv (e.g. CO spawn_mad_min = 0.164, rear_mad_min = 0.03, rear_mad_max = 40) are required to match those WSGs.

For ADMS the model is cw (channel width) so MAD doesn't apply. Phase 2 unblocks the rest of the province.

Proposed Solution: Phase 2 of #113

1. Enrich segmented streams with mad_m3s

R/frs_network_segment.R already calls frs_col_join to add channel_width from whse_basemapping.fwa_stream_networks_channel_width joined on linear_feature_id. Add a parallel call:

frs_col_join(conn, to,
  from = "fwa_stream_networks_discharge",
  cols = c("mad_m3s"),
  by = "linear_feature_id")

Verify the source table exists in fwapg (whse_basemapping.fwa_stream_networks_discharge or similar).

2. Add mad predicate to the rule evaluator

In R/frs_habitat_classify.R, extend the rule evaluator from #113 to support:

- mad: [0.164, 9999]

Translates to s.mad_m3s >= 0.164 AND s.mad_m3s <= 9999.

3. Update CSV defaults if needed

The CSV-only species classification path (for species not in the rules file) currently builds spawn_cond/rear_cond from params$ranges$spawn$mad_m3s and similar. fresh 0.11.0 already parses MAD ranges in frs_params() (verified — params$CO$ranges$spawn$mad_m3s is set to c(0.164, 9999)) but frs_habitat_classify() doesn't apply them. Either:

• A: Apply MAD predicates in the CSV-only path too
• B: Require MAD-using species to be in the rules file

Option A is more transparent (CSV row "just works"). Option B is simpler implementation.

Test approach

Use a MAD-model WSG (BULK or LDEN) sub-basin. Run frs_habitat with a rules file that uses MAD predicates for CO/CH/SK/ST. Assert MAD-filtered classification matches expectations.

For ADMS (cw model) this should be a no-op — MAD predicates not used.

Dependencies

• Habitat eligibility rules format for multi-rule species (Path 1) #113 Phase 1 (rule format and evaluator)
• Verify whse_basemapping.fwa_stream_networks_discharge schema in fwapg

Relates to

• Habitat eligibility rules format for multi-rule species (Path 1) #113 Phase 1 — depends on this for the evaluator to extend
• bcfishpass v0.5.0 — reference for which species/WSGs use MAD model