feat: add stable handles for artifact reversions

[Tech-Circle-48]

Which issue or RFC does this PR close?

Closes #1863

Rationale for this change

Artifact IDs are generated and are not convenient for users or integrations to remember and reuse. This change introduces human-readable, scope-local stable handles that resolve to one exact artifact revision.

What changes are included in this PR?

• Add the StableHandle domain model with canonical key validation.
• Add persistent mappings from (scope, family, key) to an exact ArtifactRef.
• Add SQLite and MySQL migrations for the pc_artifact_handles table.
• Add public HTTP endpoints for binding and resolving artifact handles.
• Add corresponding Python client methods and generated API bindings.
• Return explicit errors for missing handles, missing target artifacts, and conflicting rebinding attempts.
• Preserve the originally bound revision when newer artifact revisions are created.
• Update database restore documentation for the new table.
• Add model, persistence, public API, client, schema, and contract tests.

Are there any user-facing changes?

How was this change tested?

• make test
  • 4616 passed, 313 skipped, 4 deselected
• make check
  • All formatting, linting, type checking, and integration manifest checks passed
• make contract-test
  • 49 passed
• Added tests covering:
  • stable handle validation and length limits
  • binding and resolving exact artifact revisions
  • idempotent binding
  • conflicting rebinding
  • scope isolation
  • missing handles and target artifacts
  • fixed-revision resolution after a newer revision is created
  • public HTTP and Python client behavior
  • SQLite and MySQL schema definitions

AI usage statement

[hidb4ai]

ci failed

[hidb4ai]

One blocking startup issue remains in the current commit.

[hidb4ai]

Re-reviewed c3e27e0. The previous import blocker is fixed. Public HTTP/SQLite probes passed for idempotent binding, conflicting rebinding, missing targets, revision pinning, and readback after restart. One Desktop contract-generation issue remains below. Real MySQL/OceanBase execution was not verified.