Link and/or unlink evidence items to/from an intent. Dual-writes the intent_spec_evidence join table and the linked_intent_ids array on evidence_items.

evidenceIds in the response is re-fetched from intent_spec_evidence after applying link/unlink, so it reflects ALL currently-linked evidence (not just what was changed). link/unlink counts only increment for IDs that exist in the workspace. Cookie-session auth requires ?workspace_id= and a workspace membership. Claim Trust Loop (slice A): claims lists every claim the intent makes (objective, each outcome, each constraint, each edge case) with key (the anchor key), kind, claimText, the deterministic state (backed when any anchored item is backing; else unreviewed; else assumption; else none), per-tier counts, and items (the exact anchored evidence with tier, origin, source, and reviewedBy / reviewedAt, which are populated from the latest attributed review; null for legacy rows reviewed before attribution existed, and a reviewer is never inferred). general lists intent-level links anchored to no claim; they never make a claim backed. evidenceAnchors is the current anchor map and canWrite is true only for a signed-in writer.