RPC probe: fix spawn race and silence empty-body spam (win.28)

Two bugs surfaced by win.26's RPC liveness probe:

- Race in ConsoleDashboard::render: _lastRpcProbe was only written at
  the end of probeRpcServer(), so every render in the window between
  spawn and completion (~1ms, but rendered every 500ms) spawned a new
  probe thread. Result: tens of TCP connects per second to the RPC
  port. Now set _lastRpcProbe before spawning, under the same mutex.

- Dead early-return in Server::handleConnection: checked e.getMessage()
  == "empty", but DigiByteException's ctor wraps non-JSON messages in
  "Error during parsing of >>...<<", so the check never matched and
  every probe connection printed a DEBUG line. Now matches the wrapped
  form as well.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: chopperbriano

CMakeLists.txt

@@ -40,7 +40,7 @@ SET(PATCH_VERSION 0)
 SET(SO_VERSION    0)
 
 # Windows port build number (increment for each Windows-specific release)
-SET(WIN_BUILD 27)
+SET(WIN_BUILD 28)
 
 # Add source directory
 include_directories(src)

src/ConsoleDashboard.cpp

@@ -369,10 +369,14 @@ void ConsoleDashboard::render() {
     // RPC::Server that becomes dangling if the detached accept-loop thread
     // dies, so trusting getRpcServerIfSet() can show a stale port for a
     // service that isn't actually listening.
+    // Update _lastRpcProbe BEFORE spawning, otherwise every render in the
+    // ~1ms window before the probe thread finishes will spawn a duplicate.
     {
+        std::lock_guard<std::mutex> lock(_rpcProbeMutex);
         auto rnow = std::chrono::steady_clock::now();
         auto rElapsed = std::chrono::duration<double>(rnow - _lastRpcProbe).count();
         if (rElapsed >= 30.0 || !_rpcProbed) {
+            _lastRpcProbe = rnow;
             std::thread([this]() { probeRpcServer(); }).detach();
         }
     }

src/RPC/Server.cpp

@@ -208,9 +208,15 @@ namespace RPC {
                 log->addMessage("RPC call #" + std::to_string(callNumber) + " method: " + method, Log::DEBUG);
                 response = handleRpcRequest(request);
             } catch (const DigiByteException& e) {
-                if (e.getMessage() == "empty") return; // stub accept — no real connection
-                log->addMessage("Expected exception in RPC call #" + std::to_string(callNumber) + ": " + e.getMessage(), Log::DEBUG);
-                response = createErrorResponse(e.getCode(), e.getMessage(), request);
+                // Stub / probe connections: DigiByteException's ctor wraps the
+                // raw message in "Error during parsing of >>...<<" because it
+                // tries to JSON-parse the message, so match on the wrapped
+                // form. Dashboard's RPC liveness probe (TCP connect + close)
+                // produces this on every check.
+                const std::string& em = e.getMessage();
+                if (em == "empty" || em.find(">>empty<<") != std::string::npos) return;
+                log->addMessage("Expected exception in RPC call #" + std::to_string(callNumber) + ": " + em, Log::DEBUG);
+                response = createErrorResponse(e.getCode(), em, request);
                 error = true;
             } catch (const std::exception& e) {
                 log->addMessage("Unexpected exception in RPC call #" + std::to_string(callNumber) + ": " + e.what(), Log::DEBUG);