docs: document deadlock detector mutex design rationale (#744) (#4008)

docs: document deadlock detector mutex design rationale

Add comment explaining why std::sync::Mutex is used instead of
tokio::sync::Mutex in the deadlock detector.

Refs #744

Co-authored-by: houseme <housemecn@gmail.com>

crates/io-core/src/deadlock_detector.rs

@@ -117,6 +117,12 @@ impl Default for DeadlockDetectorConfig {
 }
 
 /// Deadlock detector.
+/// Deadlock detector using wait-for graphs.
+///
+/// Uses `std::sync::Mutex` (not `tokio::sync::Mutex`) because:
+/// - Locks are never held across `.await` points
+/// - Critical sections are sub-microsecond (single HashMap operations)
+/// - `tokio::sync::Mutex` would add unnecessary overhead for these short operations
 pub struct DeadlockDetector {
     /// Configuration.
     config: DeadlockDetectorConfig,