Fix select didn't work with pipes and the timeout argument

[ktock]

This PR fixes an issue that the timeout argument for the select syscall wasn't applied to pipes and it returned immedeately instead of waiting for events.

Although the select syscall passes the timeout value to the poll method of the stream implementations, this can't handle multiple fds because a fd can't notify readiness while another is blocking in poll. As a result, using the select syscall with a combination of pipes and other streams (e.g. PTY) can be problematic.

To address this, this PR implements a callback-based event notification. Each stream implementation's poll invocation receives a callback, allowing it asynchronously notify the select syscall when it becomes ready. The select syscall blocks until one of these callbacks is triggered or the timeout expires. This behviour is enabled only when PROXY_TO_PTHREAD is enabled to avoid blocking the main worker.

To maintain compatibility with non-pipefs streams, the select syscall allows stream implementations to ignore the callback and synchronously return the event status instead. In such cases, the select syscall still updates the flags accordingly.

[sbc100]

Thanks for the PR @ktock. This looks like a fairly large and complex PR, and I'm out sick right now I'm afraid so I'm not sure when I will have time to go through it.

CC'ing @atrosinenko who wrote the original pipefs back in #4935.

Also @kripken and @tlively who might be interesting the multithreaded event stuff.

[ktock]

Rebased. Could we move this forward?

[kripken]

I am unfortunately not familiar with this code (or with native pipes either). It looks like the PIPEFS code began in #4378 / #4935 by @cynecx and @atrosinenko - would one of you be interested to review this perhaps?

[ktock]

@sbc100 @kripken

This looks like a fairly large and complex PR

I've just simplified the patch by removing changes in libpthread.js. Instead, this patch uses emscripten_proxy_sync_with_ctx to synchronize the thread worker and the main worker. This patch allows pipefs and setTimeout to unblock the worker by calling emscripten_proxy_finish when an event occurs. I've added some wrapper functions for the proxying-related APIs so that the JS implementation of newselect can use them.

[ktock]

The failure in ci/circleci: test-mac-arm64 looks like unrelated to this PR:

This job was rejected because the resource class is unavailable

[sbc100]

Circle CI should be fixed in #25955

[ktock]

There seem to be a couple of issues with the tests here.

Firstly I've seen some new flakes which I guess are timing related?

Secondly they seems to be failing under lsan/asan:https://logs.chromium.org/logs/emscripten-releases/buildbucket/cr-buildbucket/8695064874929935185/+/u/Emscripten_testsuite__ASan_/stdout

emcc: error: -pthread + ALLOW_MEMORY_GROWTH may run non-wasm code slowly, see WebAssembly/design#1271 [-Wpthreads-mem-growth] [-Werror]
[240/1180] None
None
test_select_blocking (test_core.asan.test_select_blocking) ... FAIL

maybe an easy fix?

Sorry for the slow reply, and thank you for the follow-up PRs. I've also seen the ongoing PR #25990 and added comments there as well. Please let me know if there is something I can do for these features.