[SYCL][UR] Fix CUDA and HIP semaphore and image copy event recording (#21075)
This commit fixes event recording for the following functions in the
CUDA and HIP UR adapters:
- urBindlessImagesImageCopyExp
- urBindlessImagesWaitExternalSemaphoreExp
- urBindlessImagesSignalExternalSemaphoreExp
Previously, start was not called on the event and the return value of
record was ignored.
The test that should benefit from this fix is
https://github.com/intel/llvm/blob/sycl/sycl/test-e2e/bindless_images/vulkan_interop/unsampled_images_timeline_semaphore.cpp.
My own test app runs the copy operation for 100 times, and without the
fix from this PR, a race condition can be seen about 3 in 100 times. So
it may take a while to see the issue in
`unsampled_images_timeline_semaphore.cpp`, which only runs the copy
operation once.