test(integration): TestNATSBackend_EndToEnd flakes in CI (no event in 30s) #768

Problem. TestNATSBackend_EndToEnd (tests/integration/mq_nats_test.go) failed in CI with:

mq_nats_test.go:214: no event with "\"a1\"" within 30s
--- FAIL: TestNATSBackend_EndToEnd (31.34s)

Evidence (measured).

• The Integration tests (app) job of CI run 37692261415 (2026-10-07), on PR fix(build): pin Go toolchain to go.mod and show pnpm install errors #757. That PR changes the Go toolchain pin and pnpm error output, not the ingest or NATS paths.
• The failing job passed on a rerun of the same commit.
• The suite ran 276 tests in 166 s with one failure.
• Since ci(integration): split the suite into parallel parts, select by tag #756, this test calls t.Parallel() (mq_nats_test.go:170), so it now overlaps other container-backed tests in the app part.

Context (inferred, not confirmed).

• The test waits up to 30 s for an ingested row to surface as an event. Under -race with -parallel 8 on a 4-vCPU runner, the end-to-end path (NATS JetStream, ingest worker, ClickHouse, event delivery) may need longer than that when it shares the machine with other parallel tests.
• The review of ci(integration): split the suite into parallel parts, select by tag #756 named timing-sensitive integration tests as a flake risk under the new parallelism. This failure is consistent with that, but one occurrence doesn't establish the cause.

Related: #756, #743. Part of #740.