Add an agent skill that keeps the sink backends at feature parity

[dcoric]

GitProxy now has three sink backends implementing the Sink interface in src/db/types.ts: fs (NeDB), mongo, and postgres (in progress via #1687). Every change to one backend is a parity obligation on the others, and recent history shows how easily that slips:

• New Sink members (getRepoPushRollupsByCanonicalUrl, updateRepo, the migration hooks) had to be retrofitted onto the postgres adapter after landing for mongo/fs.
• New Repo fields (dateCreated / lastModified, feat(db): add dateCreated and lastModified to Repo #1656) were stored and bumped by mongo/fs but silently dropped by the postgres adapter's update allowlist, which would have made the feat: backfill repo dateCreated/lastModified via migration #1681 startup migration abort boot on postgres.

AI agents do a growing share of this work, so the guardrail should live where agents look: a skill in .agents/skills/ (exposed to Claude via the .claude/skills symlink and to opencode via a command stub) that encodes the parity contract and checklist:

• src/db/types.ts is the contract: any new member or field must be implemented in src/db/file, src/db/mongo, and src/db/postgres in the same change.
• Parity is semantic, not just structural: same defaults on create, same timestamp bumps, same projections and filters, same error behaviour, matching case sensitivity.
• Column allowlists must not silently drop new fields; postgres schema changes are append-only migrations.
• Per-backend unit tests plus the verification commands that catch drift (npm run check-types:server, backend test suites, the postgres integration lane).

Acceptance: sink-parity skill file plus opencode stub on the feat/postgres integration branch, applicable to any agent that reads AGENTS.md.

[dcoric]

Delivered by #1689, merged into the feat/postgres integration branch; lands on main with #1687.