runloop: the audio pace bit reads whether a write happened, not who might have written

RUNLOOP_PACE_AUDIO required runloop_is_libretro_running(), on the
theory that with the core paused nothing writes to the driver. Wrong:
audio_driver_menu_sample() feeds a frame of silence every menu
iteration the core is not running, with the mixer, menu sounds and
thumbnail video audio mixed in, through the same blocking
audio_driver_flush() funnel. Audio holds the menu loop too, and the
overlay said it did not.

Stop predicting and record the fact. AUDIO_FLAG_WROTE is set at the
three driver write sites, cleared at the top of runloop_iterate()
alongside the pace reset, and the bit reads it. Whatever path reached
the write - core, menu silence, mixer - the answer is the same.

audio/audio_defines.h

@@ -175,7 +175,16 @@ enum audio_driver_state_flags
     * The threaded pipeline's producer only waits for ring space while
     * this is set: with the consumer parked, waiting could never end.
     */
-   AUDIO_FLAG_STARTED      = (1 << 9)
+   AUDIO_FLAG_STARTED      = (1 << 9),
+   /**
+    * A driver write happened since the runloop last cleared this. Set at
+    * the write sites, whichever path reached them - core audio, or the
+    * frame of silence audio_driver_menu_sample() feeds while the core is
+    * paused with the mixer and thumbnail audio mixed in. Read by the
+    * runloop's pace record so "audio is holding the loop" is a fact
+    * about this iteration rather than a guess about who wrote.
+    */
+   AUDIO_FLAG_WROTE        = (1 << 10)
 };
 
 typedef struct audio_statistics

audio/audio_driver.c

@@ -1248,6 +1248,7 @@ static void audio_driver_flush(audio_driver_state_t *audio_st,
                      out_frames, mixer_gain, override);
             }
 #endif
+            AUDIO_FLAGS_SET(audio_st, AUDIO_FLAG_WROTE);
             audio->write(audio_st->context_audio_data,
                   audio_st->output_samples_buf,
                   out_frames * 2 * sizeof(float));
@@ -1321,6 +1322,7 @@ static void audio_driver_flush(audio_driver_state_t *audio_st,
                         mixer_gain, override);
             }
 #endif
+            AUDIO_FLAGS_SET(audio_st, AUDIO_FLAG_WROTE);
             audio->write(audio_st->context_audio_data,
                   audio_st->output_samples_int16,
                   out_frames * 2 * sizeof(int16_t));
@@ -1738,6 +1740,7 @@ static void audio_driver_flush(audio_driver_state_t *audio_st,
          output_frames       *= sizeof(int16_t);  /* Unit: bytes */
       }
 
+      AUDIO_FLAGS_SET(audio_st, AUDIO_FLAG_WROTE);
       audio->write(audio_st->context_audio_data,
             output_data, output_frames * 2);
    }

runloop.c

@@ -7933,6 +7933,7 @@ int runloop_iterate(void)
     * than path, so it is set here and holds for those early returns:
     * they do rely on it. */
    runloop_st->pace = RUNLOOP_PACE_NONE;
+   AUDIO_FLAGS_CLEAR(audio_st, AUDIO_FLAG_WROTE);
    if (     settings->bools.video_vsync
          && !(input_st->flags & INP_FLAG_NONBLOCKING)
          && !(runloop_st->flags & RUNLOOP_FLAG_FORCE_NONBLOCK))
@@ -8287,14 +8288,14 @@ end:
     * puts the driver into non-blocking mode for a few frames while
     * audio_sync stays true, and during those frames audio is not
     * holding anything. */
+   /* ...and a write actually happened this iteration. Not a predicate
+    * on who might have written: the menu feeds a frame of silence
+    * through the same blocking funnel while the core is paused, with
+    * the mixer and thumbnail audio mixed in, so "libretro running" was
+    * the wrong test. The flag is set at the write sites. */
    if (     (AUDIO_FLAGS_GET(audio_st) & AUDIO_FLAG_ACTIVE)
          && !(AUDIO_FLAGS_GET(audio_st) & AUDIO_FLAG_NONBLOCK)
-         /* ...and the core is actually writing samples this frame. In
-          * the menu with the core paused the driver stays in blocking
-          * mode but nothing calls audio_driver_write(). Same predicate
-          * the RUNLOOP_STATE_MENU throttle uses for this question. */
-         && runloop_is_libretro_running(runloop_st,
-               settings->bools.menu_pause_libretro))
+         && (AUDIO_FLAGS_GET(audio_st) & AUDIO_FLAG_WROTE))
       runloop_st->pace |= RUNLOOP_PACE_AUDIO;
    /* Mirrors the gate at the video_driver_scanline_after_frame() call
     * site, which skips the wait under fast-forward. A failed