[util] Improve ticket lock implementation

Atomic fetch-and-add on unlock is not needed since no other thread can
modify the serving counter after the calling thread acquired the lock.
May slightly improve performance in games relying on ID3D10Multithread.

src/util/sync/sync_ticketlock.h

@@ -26,7 +26,8 @@ namespace dxvk::sync {
     }
 
     void unlock() {
-      m_serving.fetch_add(1, std::memory_order_release);
+      uint32_t serveNext = m_serving.load() + 1;
+      m_serving.store(serveNext, std::memory_order_release);
     }
 
   private: