fix(observability): emit a finish for every cucumber test/hook to prevent orphaned runs

[xxshubhamxx]

Problem

With Nightwatch + cucumber and Test Observability enabled, a test or hook can register a start but never emit a matchable finish, so the backend leaves it running and an inactivity reaper eventually flips it (and the whole build) to timeout — inflating the reported build/execution time (observed ~4h vs ~45min of real work).

Root cause: in nightwatch/globals.js the cucumber _tests map was keyed by the non-unique testCaseId (the pickle/scenario id), not the per-attempt testCaseStartedId. On reruns, cucumber retries, and parallel/interleaved emission, a second TestCaseStarted overwrites _tests[testCaseId] and the first TestCaseFinished's delete leaves the later attempt with no entry, so its TestRunFinished is silently skipped. Hook finishes early-return when their map entry is missing, and nothing reconciles still-open runs at teardown.

Changes

1. Rerun-safe correlation — key the cucumber _tests map by the unique testCaseStartedId (per attempt) instead of testCaseId, so retries/parallel attempts no longer clobber each other. If the entry is still missing at finish time, reconstruct minimal metadata and still emit TestRunFinished (never silently drop). (The native path's getUUID also falls back to the most-recent open run instead of returning null.)
2. Complete hook classification — getCucumberHookType always returns a known hook type (classifies BEFORE_ALL/AFTER_ALL; never undefined).
3. Teardown sweep — in both after() and afterChildProcess(), before the request queue is drained, emit a synthetic finish (result failed) for any started-but-unfinished test/hook so nothing is left running. Idempotent and fully guarded (logs only; never throws or blocks the user's run).

Review hardening (follow-up commit): duplicate TestCaseFinished events are now idempotently ignored (a second finish for the same testCaseStartedId is a no-op); unmatched finishes are left to the teardown sweep rather than minting a fresh synthetic uuid mid-run, which avoids creating ghost entries in the backend.

Verification

• Unit tests added: two TestCaseStarted sharing a testCaseId but different testCaseStartedId now both produce a TestRunFinished (fails on the old keying); the teardown sweep finishes left-open _tests/hooksMap/activeTestRuns; getCucumberHookType never returns undefined. All pass; full suite green; lint/build clean.
• An event-replay driving the actual registerEventHandlers with real cucumber event shapes confirms the pre-fix code orphans a run under interleaved-attempt ordering and under a dropped finish at teardown, while the patched code emits a finish in both cases.

Notes

• All instrumentation paths remain inside try/catch (graceful degradation): the fix cannot break a user's test run.
• No new dependencies.

Versioning

Recommended bump: patch → 3.11.0 → 3.11.1

This PR carries a fix: conventional-commit prefix, corrects broken behavior (orphaned runs inflating build duration), adds no new public API surface, and introduces no breaking change. The repo's established pattern is consistent: every fix-prefixed PR maps to a PATCH bump (e.g. v3.10.2 ← PR #53, v3.10.1 ← PR #52), while new-feature PRs map to MINOR. Release is manual per RELEASE.md.

Release notes draft

What's Changed

• fix(observability): emit a finish for every cucumber test/hook to prevent orphaned runs by @xxshubhamxx in fix(observability): emit a finish for every cucumber test/hook to prevent orphaned runs #56

Full Changelog: v3.11.0...v3.11.1