RequestTagMulti should expose CUDA stream or synchronize device

[pentschev]

Currently RequestTagMulti doesn't take into consideration CUDA synchronization which is an error. The user should be able to consume any received data or release any sent data immediately after the request completes, without having to care about synchronization. Another valid approach would be to let the user specify a CUDA async stream, and in that case release the synchronization requirement back to the user.

When this is fixed, remove

ucxx/python/distributed-ucxx/distributed_ucxx/ucxx.py

Lines 376 to 377 in 7af9f19

	# TODO: We don't know if any frames are CUDA, investigate whether
	# we need to synchronize device here.