test: fix BroadcastChannel leakage in worker.js test suite

[krajorama]

What

Follow-up to #806. While reviewing that PR, test/workerTest.js's "aggregates worker responses in thread id order" and two shutdown() tests turned out to only pass by accident: jest.resetModules() creates a fresh WorkerRegistry module instance per test, each with its own ANNOUNCEMENT_CHANNEL and per-thread BroadcastChannel. Those channels keep receiving messages until explicitly closed — even once nothing in JS still references them — so a previous test's listeners stayed alive and kept reacting to later tests' messages on the same channel names.

That leakage is exactly what let those three tests pass at all: their discovery promise, which waits for a non-primary ANNOUNCEMENT echo, only ever resolved via a stray listener left over from an earlier test — never from anything the test itself set up. Run any of the three in true isolation (-t "...") against main and they fail outright (timeout or InvalidStateError: BroadcastChannel is closed).

How

• Add a guarded WorkerRegistry.resetForTesting() (throws outside test/development, mirroring the existing guard on cluster.js's scanListeners) that closes ANNOUNCEMENT_CHANNEL, the per-thread channel, and every registered worker's channel, then resets workers/requests/historicMetrics/listenersAdded.
• Call it everywhere a test loads a fresh module instance via jest.resetModules(), including a 30-iteration loop that previously leaked 30 channels per run and closed none.
• Replace the discovery/echo pattern in the two affected tests with polling the already-existing WorkerRegistry.workerCount() — a real signal, rather than hoping to observe an echo that a self-posted BroadcastChannel message never actually delivers back to its own sender.

Test plan

• npx jest — full suite passes (581/581 on current main)
• npx eslint . && npx prettier . --check && npx tsc --project . — all clean
• Previously-flaky tests pass in the full file run (30/30 across repeated runs)
• Previously-flaky tests pass in true isolation too (-t "...", 5/5 each) — this is the actual proof the fragility is gone, not just hidden
• The "Jest did not exit... asynchronous operations" open-handle warning that showed up intermittently before this fix no longer appears

Signed-off-by: György Krajcsovits gyorgy.krajcsovits@grafana.com

🤖 Generated with Claude Code

[github-actions]

Thanks for opening this pull request! Each pull request require an update in the CHANGELOG. Please update it based on your changes.