test(pool): cover cancellation during bounded acquisition

[QwQBiG]

Summary

Motivated by reviewing OpenDAL's proposed Asyncband migration in apache/opendal#8219. Its Memcached connection acquisition races pool.get() against a ten-second timeout, so an unfinished acquisition can be cancelled.

Add bounded pool tests for cancellation:

• while waiting for capacity;
• after notification but before the acquisition is polled again;
• during object creation, with another acquisition waiting.

These complement the existing recycling and replenishment cancellation tests by checking acquisition cancellation through the public pool API. They verify waiter progress, capacity recovery, and object reuse.

The tests use controlled polling without network calls or wall-clock timeouts. They check wake-driven progress without requiring FIFO notification order. Production code is unchanged; this is coverage for the existing cancellation contract, not a fix for an observed downstream failure.

Validation

• 23 pool integration tests passed on stable and Rust 1.86.
• cargo x test: 476 tests and doctests passed.
• Targeted Clippy, rustfmt, typos, and Hawkeye checks passed.

Validation was performed locally in my WSL Linux.

[QwQBiG]

This is a non-blocking, test-only follow-up and can wait until after the 0.7.2 release.

[tisonkun]

Thanks for your contribution @QwQBiG! LGTM.

Merging ...

[tisonkun]

can wait until after the 0.7.2 release.

Need not to wait for release. I generally consider our release model as a release train model where if a release failed, we start a new release from the new latest commit.

[QwQBiG]

release train model where if a release failed, we start a new release from the new latest commit.

Got it~ Thanks for the clarification and the merge~