test: frame output thread keeps accepting work after dispose()

[dmandrade]

What

On Android, every CameraFrameOutput owns a single-thread executor (IdentifiableExecutor), and HybridFrameOutput.dispose() / HybridDepthFrameOutput.dispose() only call clearAnalyzer(). Nothing calls shutdown() on it, so each disposed output leaves a live com.margelo.camera.frame thread behind. Combined with useFrameOutput never disposing its output, the count grows by one per camera mount: 1 to 8 over 8 open/close cycles in our app, and 42 before the app froze in a long session.

Test

Adds one Harness test that creates a frame output, runs a task on its thread, disposes the output, and asserts the thread rejects new work. It is red on main because the executor keeps running.

No camera session is needed: the executor thread starts on the first task.

Notes

• On iOS the thread is a DispatchQueue, which doesn't leak an OS thread but would also keep accepting work, so this may be red on iOS too. Whether "a disposed output's thread rejects work" should hold on both platforms is your call.
• The fix we run locally (patch-package) shuts the executor down in both dispose() methods and ignores late CameraX callbacks. For this test to pass, runOnThread would also have to throw after shutdown. Happy to push that here if this direction looks right.
• Related: fix: Clear native output callbacks in output hooks on unmount #4192 (clears hook callbacks on unmount; doesn't stop the thread), fix(android): Destroy the session lifecycle owner on reconfigure #4186 (Android session lifecycle owner leak).

[vercel]

@dmandrade is attempting to deploy a commit to the Margelo Team on Vercel.

A member of the Team first needs to authorize it.