GPU Metal: wait for other streams in a command buffer of its own

StreamWaitForEvents encoded the waits into the stream's open command buffer,
after whatever work had accumulated there. With the merger's SectorRefit on
all streams, stream 0's buffer (its own 4 sectors of refit) then waited for
the 7 other streams: on pbpb30 its GPU work ended 36 ms before the wait was
satisfied, and the buffer stayed incomplete meanwhile. Commit the pending
work first, so that every GPU-side wait sits at the start of a buffer with no
work before it. The host drains the stream right after the waits anyway.

pbpb30, M1 Max, 3 rounds of 10 events: 958.8 -> 959.7 ms/ev; 6 rounds of
the same change as a runtime switch: 971.2 -> 961.3 ms/ev. Metal dumps
byte-identical on pbpb30, sim5pp identical to the CPU.

Author: ktf

GPU/GPUTracking/Base/metal/GPUReconstructionMetal.mm

@@ -491,6 +491,11 @@ MetalMemRef metalResolve(const GPUReconstructionMetalInternals* internals, const
 
 void GPUReconstructionMetal::StreamWaitForEvents(int32_t stream, deviceEvent* evList, int32_t nEvents)
 {
+  // Encoded after the stream's pending work, the waits would keep its command buffer
+  // from completing until the other streams are done: wait in a buffer of its own.
+  if (mInternals->streamHasWork[stream]) {
+    CommitStream(stream);
+  }
   // Encode commands to wait for all the events
   id<MTLCommandBuffer> cb = mInternals->commandBuffers[stream];
   for (int32_t ei = 0; ei < nEvents; ei++) {