fix(mining): make XmrigManager's thread count atomic

The requested thread count is written by generateConfig() on the
mining-control thread and read on the UI thread (the thread-grid sync
under idle thread scaling, the pool auto-balance restart) without any
synchronisation. The write also stored the unclamped count first and
clamped it afterwards, so a reader could see more threads than cores.
It is now an atomic, set by one store of the clamped value.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/daemon/xmrig_manager.cpp

@@ -173,9 +173,10 @@ bool XmrigManager::generateConfig(const Config& cfg, const std::string& outPath)
 
     int hw = (int)std::thread::hardware_concurrency();
     if (hw < 1) hw = 1;
-    // Use explicit thread count (not just a hint)
-    threads_ = (cfg.threads > 0) ? cfg.threads : std::max(1, hw / 2);
-    if (threads_ > hw) threads_ = hw;
+    // Use explicit thread count (not just a hint). One store: the UI thread reads it (getRequestedThreads).
+    int t = (cfg.threads > 0) ? cfg.threads : std::max(1, hw / 2);
+    if (t > hw) t = hw;
+    threads_ = t;
 
     json j;
     j["autosave"]   = false;
@@ -324,7 +325,7 @@ bool XmrigManager::start(const Config& cfg) {
         return false;
     }
     DEBUG_LOGF("[INFO] XmrigManager: config written to %s (API port %d, threads %d)\n",
-           cfgPath.c_str(), api_port_, threads_);
+           cfgPath.c_str(), api_port_, threads_.load());
 
     // Spawn process
     if (!startProcess(binary, cfgPath, threads_)) {