Performance Improvements and Refactoring: Replace flume with crossbeam SegQueue

[fereidani]

Hi,
This PR improves performance in various parts of the code and includes some refactoring.

Main changes are:

1. In the iouring driver, I removed two heap allocations, which simplifies the implementation and should slightly improve performance. (will send it in separate PR with correct Pin usage)
2. I removed flume. As the creator of the kanal, I respect the flume project as a safe channel implementation, but in this library, the channel structure is unnecessary and not required since it is effectively being used as a queue. Queues always outperform channels because they are simpler structures with fewer features. Using the lock-free crossbeam unbounded SegQueue should improve performance under high contention, as the use case does not require blocking recv, which is a channel feature.
3. Slight refactoring and improvements in other parts of project.

[CLAassistant]

All committers have signed the CLA.

[Copilot]

Pull Request Overview

This PR aims to improve performance by removing heap allocations in the iouring driver and replacing the flume channel with crossbeam_queue::SegQueue for better lock-free performance under high contention. However, there are critical memory safety issues that must be addressed before merging.

Key Changes:

• Replace flume channel with crossbeam_queue::SegQueue for cross-thread waker communication
• Attempt to eliminate heap allocations in iouring driver by replacing Box::leak with UnsafeCell fields
• Update naming from waker_sender to waker_queue throughout the codebase

Reviewed Changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
monoio/tests/tcp_connect.rs	Removes unused LazyLock import
monoio/src/utils/ctrlc.rs	Changes memory deallocation from Box::from_raw to drop_in_place (introduces memory leak)
monoio/src/runtime.rs	Replaces flume Sender with Arc in waker cache
monoio/src/driver/uring/mod.rs	Replaces Box::leak allocations with UnsafeCell fields and switches to SegQueue (introduces critical memory safety bug)
monoio/src/driver/thread.rs	Updates global registry functions to use SegQueue instead of flume Sender
monoio/src/driver/legacy/mod.rs	Updates legacy driver to use SegQueue instead of flume
monoio/Cargo.toml	Removes flume dependency and adds crossbeam-queue

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ihciah]

Thanks for the improvement!
Concern: Switching from flume channels to Arc<SegQueue> removes the “receiver dropped → send fails” semantics. SegQueue has no close signal, and Context caches the Arc, so even after unregister_waker_queue other threads can still push into a queue that will never be drained. If the target runtime/driver is dropped, this can lead to unbounded accumulation (potential memory/task leak), which is a behavioral regression vs the previous flume sender.
Consider caching Weak and upgrade before pushing, adding a generation/alive flag, or preserving channel semantics. Does this make sense?
BTW, is there any rough reference data regarding the performance improvement when switching from channels to queues? Thank you!