[nova-loop]

What changed

PATCH /memories/:memoryId no longer advances updated_at when the values sent match what is stored. SharedMemoriesService.update() now reads the row first and builds its SET clauses with applySharedMemoryUpdates. That helper was already in shared-memory.mapper.ts (same shape as applyAgentMemoryUpdates) but nothing called it. When nothing differs it returns the stored row untouched.

• Tags are compared after normalizeTagsOnWrite, so {"tags":["DUP"]} against a stored ["dup"] is no change, and so is {"tags":null} when no tags are stored.
• A missing memory is now reported before anything is written, as the other services do, instead of being inferred from changes === 0 after the UPDATE.
• The doc comment on update() now describes the read-then-compare behaviour. The old one talked about avoiding a prior read.

Why

Closes #309

I took option 2 from the issue. The helper already existed, so this is the smaller diff here, and the two memory services now update the same way. The read-then-write race the old comment mentioned can't happen with synchronous better-sqlite3, as the issue explains.

How it was tested

• New unit cases in shared-memories.service.spec.ts under update: identical content keeps updatedAt; ["DUP"," spaced ","dup"] against stored ["dup","spaced"] keeps updatedAt; tags: null with no stored tags keeps updatedAt; one changed field among identical ones still advances it. I ran the first three against the old update() and all three failed.
• New e2e case in test/shared-memories.e2e-spec.ts: a PATCH with the stored content and differently-cased tags returns the original updatedAt.
• In apps/api: npm test (1893 passed), npm run test:e2e (822 passed), npm run lint, and npm run build all pass. I did not run the root workspace build.
• prettier --check still flags test/shared-memories.e2e-spec.ts, but only for lines that were already there on main (The format check has never run in CI, and 92 committed files no longer pass it #246). The lines I added are formatted.

Anything left out, or worth a second opinion

#311 also touches SharedMemoriesService (it changes the not-found error to MEMORY_NOT_FOUND). Whichever PR lands second will need a small rebase. The 404 here uses the same NotFoundException as findOne, so adopting #311's code is a one-line change.

Checklist

• npm run lint -w @glassbeetle/api passes (API changes)
• npm test passes (API workspace)
• npm run build passes (API workspace)
• Tests cover the change: failing-first unit tests plus one e2e case
• Docs updated if behaviour or a convention changed. Only the service doc comment changed; apps/web/README.md already describes this behaviour.
• The linked issue's acceptance criteria are met

AI assistance

This change was prepared with AI assistance. An AI coding assistant wrote the code and tests from the issue. I reviewed the diff and ran the checks above before opening the PR.