Describe the bug

The TLB WorkQueue is currently not bounded in size. It holds both shootdown requests (which are important to handle) and advance replica work requests (which are less important to handle).

Currently, if the queue is full and enqueue is called the error is ignored:

If this is uncommented, it becomes clear that some requests may be dropped if the queue is full.

Reproduction steps

1. Change the line to check for failure to enqueue (use expect to unwrap the result)
2. Run the fxmark benchmark with 96-ish cores
3. Most of the time, it will cause an error.

Expected behavior

We would like a scenario where the queue has a theoretical bound, so that we can ensure it is always possible to enqueue. This is an important property because, overall, we just want to make sure shootdowns are not lost.

Additional context

No response

Is your feature request related to a problem? Please describe.

Right now, the release_core() system call does not remove/clear an executor from that core (if there is one) so it is up to the process to make sure the core isn't executing anything before making the system call.

Describe the solution you'd like

We probably need to use an interrupt mechanism to clear the core to be released.

Describe alternatives you've considered

You could also make the system call fail (before submitting it to the log, for consistency) if there is an executor running on that core. I'm not sure what mechanism would be best for discerning if there is something running on that core.

Additional context

This feature wasn't added to the initial PR for introducing the release_core() system call due to time constraints; either potential solution shouldn't be too hard to implement.