fix(core): resolve TSAN data race in start_timeout_timer weak_ptr destruction

Release state_weak explicitly while the strong reference s is still alive
to prevent a ThreadSanitizer-detected data race on the shared_ptr control
block. Previously, the captured weak_ptr was destroyed as part of the lambda
closure after s went out of scope, causing a concurrent operator delete
(from ~weak_ptr on zero weak count) to race with the token destructor's
__release_shared read on the same control block address.

By calling state_weak.reset() before s is released, the weak ref-count
decrement occurs while a strong reference still exists, ensuring the control
block is freed deterministically on the main thread rather than racing with
the timer thread.

Fixes: ThreadSanitizer warning in enhanced_cancellation_token_test.LinkedWithTimeoutParentCancel

Author: kcenon

src/core/enhanced_cancellation_token.cpp

@@ -476,7 +476,7 @@ namespace kcenon::thread
 		std::chrono::steady_clock::time_point deadline_point) -> void
 	{
 		std::thread timer_thread(
-			[state_weak, deadline_point]()
+			[state_weak, deadline_point]() mutable
 			{
 				auto s = state_weak.lock();
 				if (!s)
@@ -544,6 +544,13 @@ namespace kcenon::thread
 				}
 
 				s->timer_active.store(false, std::memory_order_release);
+
+				// Release the weak_ptr while the strong reference (s) is still alive.
+				// This ensures ~weak_ptr() decrements the weak ref-count before the
+				// last shared_ptr is released, preventing a data race on the control
+				// block between this thread's operator delete and the token destructor's
+				// __release_shared read (ThreadSanitizer: data race).
+				state_weak.reset();
 			});
 
 		timer_thread.detach();