feat: cancel local compaction for enter staging

[WenyXu]

I hereby agree to the terms of the GreptimeDB CLA.

Refer to a related PR or issue link (optional)

What's changed and what's your intention?

PR Checklist

Please convert it to a draft if some of the following conditions are not met.

• I have written the necessary rustdoc comments.
• I have added the necessary unit tests and integration tests.
• This PR requires documentation updates.
• API changes are backward compatible.
• Schema or data changes are backward compatible.

[gemini-code-assist]

Code Review

This pull request implements cooperative cancellation for local compaction tasks in the mito2 engine, allowing EnterStaging requests to preempt running compactions instead of waiting in the pending queue. The changes introduce task ID tracking in the CompactionScheduler, add cancellation checkpoints within the compaction runner, and update the worker loop to handle cancellation notifications. Feedback identifies a critical issue in the on_compaction_cancelled logic where waiters of a cancelled task might hang if a pending compaction is immediately scheduled without notifying them first.

[gemini-code-assist]

Similar to on_compaction_finished, if there are no pending DDLs and a pending compaction is scheduled, the waiters for the cancelled compaction are not notified. This will cause the waiters to hang.

You should notify the waiters before handling the pending compaction.

Suggested change

	if self
	.handle_pending_compaction_request(
	region_id,
	manifest_ctx,
	schema_metadata_manager.clone(),
	)
	.await
	{
	return Vec::new();
	}
	if status.pending_request.is_some() {
	// The current compaction task is cancelled. We must notify its waiters.
	finish_compaction_waiters(std::mem::take(&mut status.waiters));
	if self
	.handle_pending_compaction_request(
	region_id,
	manifest_ctx,
	schema_metadata_manager.clone(),
	)
	.await
	{
	// A new compaction is scheduled. Pending DDLs (if any) will wait for it.
	return Vec::new();
	}
	}