Add a load test

load/api.ts is a k6 script run in Docker (bun run test:load), against a running API, with users signed in first (apps/api/src/load-users.ts, sessions in load/.sessions.json). docs/testing.md, "Load", has the profiles.

1. The traffic. An exported function per journey in load/api.ts, like read and write: requests over /api/v1 with the virtual user's session (user()), tagged with a name, each with a check on the status and the body. Keep the check that a user sees only their own workspace's rows when the path returns tenant data.
2. Its share of the traffic. A scenario for it in both profiles in profiles: smoke (a constant low rate for 30 s) and load (ramp(<exec>, <rate>), a fraction of TARGET_RPS).
3. The budget. Its latency threshold in options.thresholds ("http_req_duration{scenario:<name>}": ["p(95)<…", "p(99)<…"]); the failure and check rates already apply to every request. The run fails when one does; raise one only on purpose. Data the journey needs that a fresh user lacks is created through the API in the function itself or in setup(), never in the database.
4. Types. bun run --filter @repo/load check-types (k6's types from @types/k6).
5. Run it. The API up (bun dev, in the background), then bun run test:load (the 30-second smoke CI runs); PROFILE=load bun run test:load for the ramp to TARGET_RPS. Against another environment: BASE_URL, with that environment's variables so the sessions are made in its database.

Done when

The smoke passes locally, check-types passes, and the nightly run's target (LOAD_TARGET_RPS, docs/repository-settings.md) still holds.