RadioLib edge-triggered interrupts robustness (#9658)

* Fix potential race condition (read, ISR write, clear) in NotifiedWorkerThread

* Check for missed edge-triggered interrupts race condition between startReceive and enableInterrupt

* Occasionally poll to catch missed RX_DONE interrupts. (RadioLibInterface::pollMissedIrqs)

* Simplify RadioLibInterface::checkRxDoneIrqFlag()

---------

Co-authored-by: Ben Meadors <benmmeadors@gmail.com>

src/concurrency/NotifiedWorkerThread.cpp

@@ -76,8 +76,10 @@ bool NotifiedWorkerThread::notifyLater(uint32_t delay, uint32_t v, bool overwrit
 
 void NotifiedWorkerThread::checkNotification()
 {
-    auto n = notification;
-    notification = 0; // clear notification
+    // Atomically read and clear. (This avoids a potential race condition where an interrupt handler could set a new notification
+    // after checkNotification reads but before it clears, which would cause us to miss that notification until the next one comes
+    // in.)
+    auto n = notification.exchange(0); // read+clear atomically: like `n = notification; notification = 0;` but interrupt-safe
     if (n) {
         onNotify(n);
     }

src/concurrency/NotifiedWorkerThread.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include "OSThread.h"
+#include <atomic>
 
 namespace concurrency
 {
@@ -13,7 +14,7 @@ class NotifiedWorkerThread : public OSThread
     /**
      * The notification that was most recently used to wake the thread.  Read from runOnce()
      */
-    uint32_t notification = 0;
+    std::atomic<uint32_t> notification{0};
 
   public:
     NotifiedWorkerThread(const char *name) : OSThread(name) {}