Fix data race regarding server_quit in EditorExportPlatformWeb

[jsjtxietian]

Fixes #87995

[akien-mga]

Looks good.

@RandomShaper Maybe we need some documentation for contributors on when/how to use SafeNumeric in threads.

[RandomShaper]

Just for the records, if we had a time-outable semaphore wait function, this loop could be more similar to other ones we have in similar contexts across the codebase. As such, the exit flag could be a raw bool, as in the other cases.

Alternatively, if we had a Godotic relaxed atomic, it could have been used it, instead of SafeNumeric, which due to its acquire-release nature does more than is strictly needed here. The data race would be avoided just by ensuring the flag is read/written atomically, which generally would come for free at the assembly level.

All that said, and not being either the case, this PR does the right thing.

PS: Regarding docs, I hope to have some time at some point to write a few bullet points. Feel free to remind me in some time if I forget. 😃

[akien-mga]

Thanks!

[Faless]

I'm not sure this actually fixed the race condition.

The call to server->stop() might happen while the thread is inside ej->server->poll();. So AFAIU we should server_thread.wait_to_finish(); before calling server->stop().

[RandomShaper]

The problem here was a data race: a thread reading from a memory location being written by another. That's not generically safe, so ensuring the reads and writes to that variable are atomic is enough.