Temporary: Do not early release mutex when not ganging

It seesm the Release() function is not reliable and can cause segfaults.
This is a temporary work-around until the Release() function is fixed.

Change-Id: I95470a800c6153673e4b8f4fe46a646903325074

runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp

@@ -945,7 +945,6 @@ hsa_status_t GpuAgent::DmaCopy(void* dst, core::Agent& dst_agent,
   }
 
   ScopedAcquire<KernelMutex> lock(&sdma_gang_lock_);
-  if (gang_factor == 1) sdma_gang_lock_.Release();
   // Manage internal gang signals
   std::vector<core::Signal*> gang_signals;
   if (gang_factor > 1) {