perf(ffe): fix FEATURE_FLAGGING_AND_EXPERIMENTATION CI regression on Node.js

[rochdev]

Summary

• FEATURE_FLAGGING_AND_EXPERIMENTATION regressed from <3 minutes to 7+ minutes in dd-trace-js CI (nodejs express4 shard).
• Two independent costs were identified from the real CI job log/timing breakdown, and both are fixed here:

1. Exposures: waiting on a periodic flush timer instead of flushing immediately

• tests/ffe/test_exposures.py's wait_for_exposure_event/wait_for_min_exposure_count helpers were purely polling interfaces.agent.wait_for(..., timeout=30) for exposure events, at the mercy of the exposures writer's periodic flush interval.
• This adds a weblog.get("/flush") call before polling, so the weblog forces its exposures writer to flush immediately via the existing /flush endpoint pattern, instead of waiting on the timer.
• Wired the Node.js side: the 4 Node weblogs (express, express4-typescript, fastify, nextjs) now publish the ffe:writers:flush diagnostics channel (already consumed inside dd-trace-js) from their /flush handlers.
• weblog.get("/flush") is a no-op for weblogs/languages that don't implement the route, so this is safe across all libraries.
• Confirmed in CI: test_exposures.py now completes in ~30s instead of stalling on the writer's timer.

2. EVP flagevaluation: missing_feature tests were still executing under xfail, burning their full timeout

• missing_feature (and bug, etc.) map to pytest.mark.xfail, not skip (utils/_decorators.py) — the test body still runs. dd-trace-js has no implementation of /api/v2/flagevaluation at all, so tests/ffe/test_flag_eval_evp.py's 8 non-@scenario_crash classes each ran to completion and burned their full 30s wait_for timeout before failing as expected. That's ~240s, the actual dominant cost of the scenario.
• Adds @slow (existing skip_if_xfail mechanism, already used on 3 sibling classes in this file via @scenario_crash) to the remaining 8 classes. @slow only converts a test to a hard skip when it's paired with the manifest's declaration marker for the current library — so this only skips for libraries where the feature is currently declared missing (today: Node.js for all 8, plus golang for the 3 ObserveFullData_* classes it doesn't yet implement).
• Self-healing: once a library implements the feature and its manifest entry is removed, @slow becomes a no-op and the test runs for real. Confirmed against golang's actual CI timing (the one library that implements this today) that real event delivery is on the order of 1-2s, nowhere near the 30s ceiling, so no added flakiness risk anywhere.

Test plan

• Re-run the FEATURE_FLAGGING_AND_EXPERIMENTATION scenario for nodejs (express4/express4-typescript/fastify/nextjs) in CI and confirm it completes well under 3 minutes.
• Confirm tests/ffe/test_exposures.py still passes for nodejs.
• Confirm tests/ffe/test_flag_eval_evp.py still passes for golang (only the 3 ObserveFullData_* classes should skip there).
• Confirm other languages are unaffected.

🤖 Generated with Claude Code

[github-actions]

CODEOWNERS have been resolved as:

tests/ffe/test_exposures_datadog_agent.py                               @DataDog/feature-flagging-and-experimentation-sdk
tests/ffe/test_flag_eval_evp.py                                         @DataDog/feature-flagging-and-experimentation-sdk
utils/build/docker/nodejs/express/app.js                                @DataDog/system-tests-reviewers
utils/build/docker/nodejs/express4-typescript/app.ts                    @DataDog/system-tests-reviewers
utils/build/docker/nodejs/fastify/app.js                                @DataDog/system-tests-reviewers
utils/build/docker/nodejs/nextjs/src/app/flush/route.js                 @DataDog/system-tests-reviewers

[datadog-datadog-us1-prod]

✅ All CI checks and tests passed.

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: f90923f | Docs | View more details | Give us feedback!}