Fix async resolver worker deadlock

[chrmarti]

Summary

Release the in-flight map mutex before notifying async resolution waiters.

The previous if let lock(...).get(...).cloned() left the sole resolver worker unable to process a second distinct queued request, so later async callers waited forever.

Regression coverage

tests/async_resolution.rs runs two distinct async resolutions in a subprocess. The parent kills and fails a child that does not exit within five seconds, so a wedged std::thread cannot hang the whole test suite.

Verified with isolated build directories:

• fix commit: passes in ~20 ms
• old runtime dependency 4e2bfc6: fails in 5.01 s with async resolver worker did not process its second distinct request

Validation

• cargo test --features tokio --test async_resolution
• cargo test --features tokio
• cargo clippy --all-targets --features tokio -- -D warnings
• cargo fmt --all -- --check