fix(oc): reject duplicate recording titles per project (ENG-71)

[henokteixeira]

Summary

🤖 Generated with Nori

Fixes ENG-71 (Urgent). POST /api/oc/recordings silently de-duplicated by (project_id, title) — it returned the existing recording instead of creating a new one. The client then requested an upload URL for that old recording_id, whose GCS path is deterministic, so the PUT overwrote the original blob. The old recording kept its metadata but now held the new audio — silent, unrecoverable data loss.

• create_recording and update_recording now reject a duplicate title in the same project with ConflictError → 409 {detail, code: "CONFLICT"} (via the existing global handler — routers stay thin).
• Match is trimmed, case-sensitive, exact. Blank/whitespace titles normalize to NULL and are exempt. Split-derived (split_from_id set) and archived-after-split rows are exempt (they legitimately reuse a parent title).
• GET /api/oc/recordings gains an optional title filter, used by the client's save-time pre-check.

No schema change → no migration in this PR.

Known limitations / fast-follows

• Concurrency race (deferred DB hardening). The uniqueness check is service-layer only (SELECT-then-INSERT, no row lock/constraint), so two concurrent creates of the same title in one project could still both insert. This is far narrower than the original always-overwrite bug (which the service check fully closes for the reported single-user flow), but it is reachable across two devices/users. Recommended fast-follow: a partial unique index plus an IntegrityError→ConflictError backstop, gated on a pre-deploy audit (run against prod first; expect ~0 rows since the old dedup prevented exact duplicates):

  SELECT project_id, btrim(title) AS norm, count(*), array_agg(id)
  FROM oc_recordings
  WHERE title IS NOT NULL AND btrim(title) <> ''
    AND split_from_id IS NULL AND splitting_status <> 'archived_after_split'
  GROUP BY project_id, btrim(title) HAVING count(*) > 1;

  op.create_index(
      "uq_oc_recordings_project_title", "oc_recordings", ["project_id", "title"],
      unique=True, postgresql_where=sa.text("split_from_id IS NULL"),
  )

• Case-sensitivity is exercised in tests on SQLite (ASCII = is case-sensitive there) and relies on Postgres' default collation in prod; the fast-follow index makes it explicit.
• The client-side 409 UX (conflict status + rename dialog + save-time pre-check) lands in a separate oral-collector PR.

Test Plan

• uv run pytest tests/test_oc_recording_service.py — 38 passed (12 new ENG-71 cases: create/update reject, trim-on-store, case-sensitive, blank→NULL, per-project scoping allowed, split-child & archived-parent exemption, list title filter)
• Full suite uv run pytest — 665 passed, 2 skipped
• ruff check + ruff format --check + mypy clean
• Staging: create a 2nd recording with an existing title in the same project → returns 409 (not 200/silent); original blob untouched
• Staging: PATCH-rename a recording to an existing title → 409, no 500