Handle special cases when stream value is hipStreamLegacy (0x01) or hipStreamPerThread (0x02) (#343)

* Updated stream code to handle special cases when stream value is 0x01 or 0x02

* Removed extra definitions and updated tests to account for special case

* Modified stream.cpp so that each thread assigned a unique stream ID when hipStreamPerThread is used as stream value. Modified tests to check that threads are assigned unique, repeated values when hipStreamPerThread is called

* Updated idx_offset, stream_map, and thread counter to be in one struct.

* Update stream.cpp to only use add_stream() and update tests for seperate unit test for hipStreamPerThread

* Remove unecessary comment

* Removed unecessary line

* Updated tests and stream.cpp to update stream ID correctly

* Updated test structure

projects/rocprofiler-sdk/tests/bin/hip-streams/compute_comm_overlap.cpp

@@ -110,6 +110,9 @@ main()
     const int gridSizePerStream = 104;  //(int)ceil((float)elements_per_stream/blockSize);
 
     HIP_ASSERT(hipEventRecord(start));
+    // Extra copy with null stream
+    HIP_ASSERT(hipMemcpyAsync(
+        &d_input1[0], &h_input1[0], bytes_per_stream, hipMemcpyHostToDevice, nullptr));
     // split H2D copies and kernel calls into separate loops
     for(int i = 0; i < num_streams; i++)
     {