Add normalized identity-domain diagnostics for M3 set comparison

[jacksenpierce]

Status

Provisional feature hypothesis. A reported workspace incident compared canonical source pointers with local alias pointers as raw strings and allegedly produced a false uncovered-set claim; that particular database and script have not yet been independently reproduced.

Problem to investigate

External research collections may identify the same thing as a local pointer, a canonical source pointer, a node UUID, or a project-specific occurrence ID. Ad hoc string comparison across those domains can produce confident but false set reports.

Discovery questions

• What identity mappings can Parmesan authoritatively know, rather than guess?
• How should a caller declare comparison grain and identity domain?
• Can a bounded diagnostic report normalized values, unresolved IDs, duplicates, union/intersection/differences, and domain mismatches without pretending aliases are automatically equivalent?

Non-goal

Do not encode one project-specific alias convention as universal Parmesan semantics.

Related to #22.

[jacksenpierce]

Post-4.0.2 triage: M3 now supplies the canonical conceptual method, so this is no longer a request to define set algebra. The remaining product work is an operational diagnostic that requires callers to declare grain and identity domain, reports unresolved or duplicate identities, and computes comparisons only after explicit normalization.