Fix/pr 5554 rebased

[DavertMik]

Reproduction project: https://github.com/gololdf1sh/codeceptjs-plugins-test

---

Issue 1 — aiTrace: step artifacts race condition

Steps to reproduce:

1. Enable aiTrace plugin
2. Run a test that performs multiple steps
3. Check trace.md — artifacts (screenshots, HTML, ARIA) are missing or incomplete for some steps

Cause: persistStep() was called outside the recorder chain — the async capture ran in parallel with test execution, leading to race conditions.

Fix: Moved persistStep() inside recorder.add() so artifact capture is sequenced with the test flow.

---

Issue 2 — analyze: AI report silently skipped

Steps to reproduce:

1. Enable analyze plugin, set ANTHROPIC_API_KEY
2. Run tests with --ai flag where some tests fail
3. Observe: no AI report is printed

Cause: Two issues: (a) printReport() is async but was called without await — the event handler returned immediately; (b) process.exit() in run.js fires on a 1-second timeout, killing the process before the AI API response arrives.

Fix: Added await to printReport(), set CODECEPT_DISABLE_AUTO_EXIT=1 before the report, and call process.exit() explicitly after the report finishes.

---

Issue 3 — screencast: recording starts too late (misses first steps)

Steps to reproduce:

1. Enable screencast plugin with on: 'test'
2. Run a test with navigation + interactions
3. Check the recorded video — the first steps are missing

Cause: Recording was triggered on event.step.started — the first step's actions could execute before the screencast started.

Fix: Moved the recording start to event.test.started, which fires after helper._before() (page is available) but before any steps execute.

---

Issue 4 — pageInfo: plugin silently disabled (helper resolved too early)

Steps to reproduce:

1. Enable pageInfo plugin
2. Run a failing test
3. Observe: no pageInfo artifact is generated, no error message

Cause: Two issues: (a) pickActingHelper() was called at plugin init time when Container.helpers() may not be populated yet — if it returned null, the plugin silently exited; (b) recorder.add() calls lacked force=true, so they were skipped when the recorder was in a rejected state (which it is after a test failure).

Fix: Moved helper resolution inside the event.test.failed handler (lazy), and added force=true to both recorder.add() calls.

---

Issue 5 + Issue 9 — aiTrace: failed step artifacts lost / event.step.failed async handler race

Steps to reproduce:

1. Enable aiTrace plugin
2. Run a test that fails on a specific step
3. Check trace.md — the failed step has no screenshot, HTML, or ARIA snapshot

Cause: event.step.failed handler was async but Node.js EventEmitter doesn't await async handlers. The artifact capture promise was fire-and-forget — event.test.failed / persist() ran before capture finished.

Fix: Changed event.step.failed to a sync handler that stores the capture promise in pendingArtifactCapture. The event.test.failed handler awaits this promise (inside recorder.add with force=true) before persisting the trace.

---

Issue 6 — heal: steps filter in recipes ignored during code suggestions

Steps to reproduce:

1. Define a heal recipe with steps: ['click'] filter
2. Trigger healing on a fillField failure
3. Observe: the recipe is used despite the step name not matching

Cause: getCodeSuggestions() in heal.js didn't filter recipes by their steps array — it matched only by context name.

Fix: Added .filter(r => !r.steps || !stepName || r.steps.includes(stepName)) after matchRecipes() in getCodeSuggestions().

---

Issue 7 — heal: healLimit not enforced correctly

Steps to reproduce:

1. Set healLimit: 2 in heal plugin config
2. Run a test where 4+ steps fail
3. Observe: all steps are healed, limit is ignored

Cause: Two issues: (a) healTries was never reset between tests; (b) the limit check was outside recorder.catchWithoutStop — since catchWithoutStop registers a persistent handler, the outer check ran only once at registration time.

Fix: Reset healTries = 0 in event.test.before, and moved the limit check inside the catchWithoutStop callback so it's evaluated on every caught error.

---

Issue 10 — analyze: container typo (lowercase)

Steps to reproduce:

1. Enable analyze plugin with --ai flag
2. Run tests where some fail
3. Observe: ReferenceError: container is not defined in console

Cause: Line 297 used container.plugins() (lowercase) instead of Container.plugins().

Fix: Changed container to Container.