Describe private evidence inputs generically in link's code and docs

[NewGraphEnvironment]

If we do it: link's public code, docs and run stamps describe their private inputs generically, and no longer name the private repository they come from. If we never do: each new driver and doc repeats the name, and link keeps reading as a map of private infrastructure. None of it is a credential. This is about how a public repo reads, the same rule behind the host-address scrub (rtj#257).

Where the name appears on main (outside planning/ and comms/)

• data-raw/habitat_validate.R: the environment variable LNK_KNOWLEDGE_DIR, plus comments naming the repo.
• data-raw/query_habitat_thresholds_fiss.R: 7 mentions.
• CLAUDE.md: the Research: calibrate CH and BT gradient and channel-width thresholds from observations #284 status section.
• research/habitat_validation.md and research/habitat_thresholds.md: method lines, and "knowledge @ " in the reconciliation text.
• Run stamps written by the drivers (data-raw/logs/*/stamp.txt): "knowledge @ ".
• The Research: calibrate CH and BT gradient and channel-width thresholds from observations #284 issue body and the PR Region-scoped observation pooling as bundle data: lnk_species_pooling() (#290) #291 body.

The user_* CSVs' hits are the English phrase "local knowledge" in bcfishpass notes, not this.

Proposed

• Forward-only. New issues, code and docs say "the private evidence repo" or name the table, not the repository.
• Rename LNK_KNOWLEDGE_DIR → LNK_EVIDENCE_DIR, keeping the old name as a deprecated fallback for one release. Stamps then record "evidence @ ".
• Reword the current doc and comment mentions, and the two bodies.
• History (commits, planning/archive/) stays as it is.

Relates to #283, #284