fix napi_async_work prevent Node.js from exiting

[toyobayashi]

fixes #25

[toyobayashi]

seems this breaks cleanup hooks at process beforeExit event ...

[toyobayashi]

immediately call napi_cancel_async_work after napi_queue_async_work may cause error, in random probability

emnapi worker pool size in test is too great

[RReverser]

The worker ref/unref logic here seems a bit complicated TBH, especially with various checks, postMessage and so on.

Node.js doesn't care which handle keeps the process alive, so you don't need to .ref() all individual workers when running an async operation - the MessagePort from a71cc25 should be enough.

From worker side, it should be enough to only .unref() workers once when an emnapi thread is created, and leave the rest of ref/unref to that MessagePort. I think that would simplify the code quite a bit.

[toyobayashi]

Done.

[RReverser]

Nice, looks much simpler!

[RReverser]

By the way, I wanted to ask - why did you need worker = worker.worker || worker? Isn't PThread.pthreads[id] always a Worker already?

[toyobayashi]

This is for backward compatibility, I remember emscripten once refactored PThread.

emscripten-core/emscripten#17492