Adding Perfetto support (#867)

* Perfetto submodule

* include/rocprofiler-sdk/cxx/perfetto.hpp

- adapted from tests/common/perfetto.hpp
- updated json-tool to use <rocprofiler-sdk/cxx/perfetto.hpp>

* Update include/rocprofiler-sdk/cxx

- add details/delimit.hpp
- add details/join.hpp
- extend details/mpl.hpp
- extend details/operators.hpp

* Update lib/rocprofiler-sdk/hsa/async_copy.cpp

- update MEMORY_COPY direction names

* Preliminary perfetto support

* Update lib/rocprofiler-sdk-tool/generatePerfetto.cpp

- fix getting roctx msg vs. buffer operation name

* Temporary variable restructuring

* Perfetto patches after rebasing onto main

* Revert lib/rocprofiler-sdk/hsa/async_copy.cpp

- revert name

* Update lib/rocprofiler-sdk-tool/generatePerfetto.cpp

- fix ReadTrace

* Update tests/bin/hip-in-libraries

- sleep_for

* Support PFTRACE output format option in rocprofv3

* Change perfetto logging

* Update rocprofv3 tests to generate pftrace output

* Minor tweak to json-tool.cpp

* Update requirements.txt for perfetto testing

* Fix data race on amount_read in generatePerfetto.cpp

* Add testing for pftrace output

- relatively simple testing which verifies that the pftrace file has the same number of entries as JSON data for HIP/HSA/marker/kernel/memory_copy

* Fix import in perfetto_reader.py

* Fix data race in generatePerfetto.cpp

tests/bin/hip-in-libraries/hip-in-libraries.cpp

@@ -56,7 +56,7 @@ namespace
 {
 using auto_lock_t = std::unique_lock<std::mutex>;
 auto   print_lock = std::mutex{};
-size_t nqueues    = 8;
+size_t nqueues    = 4;
 size_t nthreads   = 4;
 size_t nitr       = 500;
 size_t nsync      = 10;
@@ -112,7 +112,8 @@ main(int argc, char** argv)
 
     {
         auto vector_ops_thread = std::thread{run_vector_ops, nthreads, nqueues};
-        auto transpose_thread  = std::thread{run_transpose, nthreads, nitr, nsync};
+        std::this_thread::sleep_for(std::chrono::milliseconds{100});
+        auto transpose_thread = std::thread{run_transpose, nthreads, nitr, nsync};
 
         vector_ops_thread.join();
         transpose_thread.join();