Process polling

[igankevich]

This PR adds process polling capabilities via Process type that implements Source.

• On Linux the polling is done via pidfd.
• On MacOS/BSD it is done via EVFILT_PROC filter of kqueue.
• On Windows it is done via job objects.

To implement Process I made the following changes.

• No changes to epoll-based and poll-based selectors: pidfd is like any other file descriptor.
• Minor changes to kqueue-based selector and related is_* functions to support new EVFILT_PROC filter and prevent code repetition.
• Multiple changes to Windows selector. This selector now supports registering arbitrary handles to be polled. Handle's address is used as the token (lpCompletionKey), and each selector maintains a mapping between the handles and the associated data HandleInfo. For processes the associated data consists of user-provided token. Currently only processes are registered as handles, but everything else can be re-implemented using handles to simplify the code (e.g. Arc<CompletionPort> will no longer be needed in NamedPipe and it will be possible to register the same pipe in several selectors). Please let me know if this is something you want to be done in this PR.
• tests/process.rs test.
• Changes to tests/util module to support large number of events.

Besides process polling this PR includes fixes suggested by the latest Clippy version.

EDIT

The motivation behind non-blocking process polling is to simplify projects that use mio to collect output from child processes (e.g. test harnesses, virtual networks). Currently such projects either use a separate thread to wait for child processes via wait syscall (that waits for any process to exit) or poll with small timeout and check whether the processes exited after each poll. The former needs mutexes to synchhorinze the waiting thread with the poller, and the latter is as efficient as active polling. With proper process polling such projects would use the same thread to monitor process status, this would greatly simplify their code bases.

EDIT 2

I tested the changes on Linux, MacOS, FreeBSD and Windows using the same commands as in CI and latest Rust version.

[igankevich]

Fixed (hopefully) the errors reported in CI. Had to use -Zexport-executable-symbols to make leak sanitizer work in CI (rust-lang/rust#111073).

[igankevich]

Thanks for approving the pipeline. Hopefully I fixed the flaky test :)

It seems that leak sanitizer doesn't work with panicking tests, so I disabled them when compiled under this sanitizer. The error that we get here might be related to a custom allocator. I'm not sure mio has something to do with this error. I couldn't find any Rust-related issue with the same error.

EDIT

I couldn't reproduce leak sanitizer error in the Docker container with Ubuntu 24.04 and Rust nightly. Not sure why it's happening in the CI.