lnk_presence(): move the hard-coded presence groups into the bundle #292

If we do it: the species groups that promote presence ("any salmon present means all salmon present") are bundle data, logged and hashed like every other model input. If we never do: a model input lives in code, outside config_hash, and a change to it is invisible to run provenance.

Problem

lnk_presence() defaults groups = list(salmon = c("ch","cm","co","pk","sk"), ct_dv_rb = c("ct","dv","rb")) (R/lnk_presence.R:67-70). The pipeline consumes it (R/lnk_pipeline_run.R:227) for bcfishpass-parity presence promotion, so it changes model output. #290 introduced species_groups.csv for observation pooling, but deliberately left presence alone:

• the concept differs (presence promotion vs evidence pooling);
• the groups differ (ct_dv_rb vs CHAR = BT, DV);
• folding them together would change pipeline output.

Proposed

• A bundle file, e.g. presence_groups.csv (group_code, species_code), or a role column in species_groups.csv. Whichever it is, declare it under files: so it enters config_hash.
• lnk_pipeline_run() passes it to lnk_presence(groups = ).
• bcfishpass carries today's two groups verbatim (parity); default starts identical.
• Verify byte-identical streams_access on ADMS and BULK before and after.

Relates to #290, #236, #189