Correct copy completion signal handling.

Remove agent lookup in time stamp translation for IPC signals.  The copy
agent handle is not shared so does not need to be checked for cross
process use.  Cross process copy-timestamp read is illegal and continues
to deliver garbage.

Store the copy agent properly when doing CPU-CPU copies.

Change-Id: Ib4008f66ff866922047749dd556c84a32021c1fd

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

@@ -471,6 +471,7 @@ hsa_status_t Runtime::CopyMemory(void* dst, core::Agent& dst_agent,
   // For cpu to cpu, fire and forget a copy thread.
   const bool profiling_enabled =
       (dst_agent.profiling_enabled() || src_agent.profiling_enabled());
+  if (profiling_enabled) completion_signal.async_copy_agent(&dst_agent);
   std::thread(
       [](void* dst, const void* src, size_t size,
          std::vector<core::Signal*> dep_signals,