Fix connection data race probably causing flaky tests

SilKit/source/core/vasio/ConnectKnownParticipants.cpp

@@ -161,6 +161,10 @@ void ConnectKnownParticipants::UpdateStage()
 {
     SILKIT_TRACE_METHOD_(_logger, "()");
 
+    // Serialize the whole transition: _connectStage is atomic, but the check-then-set below is not,
+    // so concurrent callers would otherwise fire the same stage callback (e.g. set_value()) twice.
+    std::lock_guard<decltype(_stageMutex)> stageLock{_stageMutex};
+
     if (_connectStage == ConnectStage::INVALID)
     {
         return;

SilKit/source/core/vasio/ConnectKnownParticipants.hpp

@@ -127,6 +127,10 @@ class ConnectKnownParticipants
 
     std::atomic<ConnectStage> _connectStage{ConnectStage::INVALID};
 
+    // Serializes the check-then-set stage transitions in UpdateStage(), which can be called
+    // concurrently from the caller thread (StartConnecting) and the IO worker thread (connect callbacks).
+    std::mutex _stageMutex{};
+
     mutable std::mutex _mutex{};
     std::unordered_map<std::string, std::unique_ptr<Peer>> _peers;