[SYCL][NFC] Introduce unit-tests for ONEAPI_DEVICE_SELECTOR

[AlexeySachkov]

There are E2E tests which are essentially testing the correctness of ONEAPI_DEVICE_SELECTOR env variable testing, but that can be done in way more efficient way which is unit-tests.

Note that this PR does not aim to cover ONEAPI_DEVICE_SELECTOR parsing with unit-tests in full, but rather aims to replace the most obvious candidates from E2E tests with unit-tests, providing some foundation for expanding unit-tests in follow-up PRs.

[AlexeySachkov]

I had a failure in the first run of unittests on Windows, which looked like:

[ RUN      ] SchedulerTest.FailedKernelExceptionException Code: 0xC00000050x00007FF68FF43D13, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x293D13 byte(s), urEventWaitRedefineCheckEvents() + 0x223 byte(s), D:\github\_work\llvm\llvm\src\sycl\unittests\scheduler\CommandsWaitForEvents.cpp, line 53 + 0x11 byte(s)0x00007FFA7CF3EAAB, D:\github\_work\llvm\llvm\build\bin\ur_adapter_mock.dll(0x00007FFA7CF30000) + 0xEAAB byte(s), driver::urEventWait() + 0x7B byte(s), D:\github\_work\llvm\llvm\build\_deps\unified-runtime-src\source\adapters\mock\ur_mockddi.cpp, line 5100 + 0x0 byte(s)0x00007FFA7A930B0E, D:\github\_work\llvm\llvm\build\bin\UR_LOADER.dll(0x00007FFA7A900000) + 0x30B0E byte(s), urEventWait() + 0x2E byte(s), D:\github\_work\llvm\llvm\build\_deps\unified-runtime-src\source\loader\ur_libapi.cpp, line 4700 + 0x7 byte(s)0x00007FF68FE5B812, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x1AB812 byte(s), sycl::_V1::detail::waitForEvents() + 0x102 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\memory_manager.cpp, line 131 + 0x1E byte(s)0x00007FF68FE5B3AA, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x1AB3AA byte(s), sycl::_V1::detail::MemoryManager::release() + 0x2A byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\memory_manager.cpp, line 260 + 0x0 byte(s)0x00007FF68FEC329A, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x21329A byte(s), sycl::_V1::detail::ReleaseCommand::enqueueImp() + 0x4FA byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\commands.cpp, line 1315 + 0x0 byte(s)0x00007FF68FEC1831, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x211831 byte(s), sycl::_V1::detail::Command::enqueue() + 0x3E1 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\commands.cpp, line 930 + 0x0 byte(s)0x00007FF68FED6B22, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x226B22 byte(s), sycl::_V1::detail::Scheduler::GraphProcessor::enqueueCommand() + 0x172 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\graph_processor.cpp, line 123 + 0x0 byte(s)0x00007FF68FED67C6, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x2267C6 byte(s), sycl::_V1::detail::Scheduler::waitForRecordToFinish() + 0x526 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\scheduler.cpp, line 84 + 0x0 byte(s)0x00007FF68FED5D63, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x225D63 byte(s), sycl::_V1::detail::Scheduler::removeMemoryObject() + 0x1F3 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\scheduler.cpp, line 281 + 0x13 byte(s)0x00007FF68FED4FFE, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x224FFE byte(s), sycl::_V1::detail::Scheduler::cleanupDeferredMemObjects() + 0x34E byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\scheduler.cpp, line 522 + 0xE byte(s)0x00007FF68FED4A53, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x224A53 byte(s), sycl::_V1::detail::Scheduler::cleanupCommands() + 0x33 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\scheduler.cpp, line 412 + 0x0 byte(s)0x00007FF68FED2567, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x222567 byte(s), sycl::_V1::detail::Scheduler::NotifyHostTaskCompletion() + 0x3A7 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\scheduler.cpp, line 471 + 0xD byte(s)0x00007FF68FEBAB16, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x20AB16 byte(s), sycl::_V1::detail::DispatchHostTask::operator()() + 0x8C6 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\scheduler\commands.cpp, line 491 + 0x10 byte(s)0x00007FF68FE18476, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x168476 byte(s), sycl::_V1::detail::ThreadPool::worker() + 0x1A6 byte(s), D:\github\_work\llvm\llvm\src\sycl\source\detail\thread_pool.hpp, line 51 + 0x0 byte(s)0x00007FF68FE13291, D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\SchedulerTests.exe(0x00007FF68FCB0000) + 0x163291 byte(s), std::thread::_Invoke<std::tuple<`sycl::_V1::detail::ThreadPool::start'::`3'::<lambda_1> >,0>() + 0x11 byte(s), C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.33.31629\include\thread, line 56 + 0x0 byte(s)0x00007FFAA0826B4C, C:\WINDOWS\System32\ucrtbase.dll(0x00007FFAA0800000) + 0x26B4C byte(s), _recalloc() + 0x5C byte(s)0x00007FFAA2D04CB0, C:\WINDOWS\System32\KERNEL32.DLL(0x00007FFAA2CF0000) + 0x14CB0 byte(s), BaseThreadInitThunk() + 0x10 byte(s)0x00007FFAA32FE8AB, C:\WINDOWS\SYSTEM32\ntdll.dll(0x00007FFAA3280000) + 0x7E8AB byte(s), RtlUserThreadStart() + 0x2B byte(s)Access violation

I'm sure that it couldn't be caused by changes in the PR, I wasn't able to reproduce it locally and it passed on the restart.

I will cancel remaining Windows E2E tests job, because those tests shouldn't be affected by the PR and they passed in the first run, logs here.

With that said, pre-commit seems to be passing for the PR and I will proceed with merge