DPCTLQueue_Memcpy, _Prefetch, _Memadvise become asynchronous

[oleksandr-pavlyk]

The return value of these functions changed. Declaration, documentation,
tests and implementations updated in dpctl-capi/.

dpctl/_backend.pxd updated for changed return type. Usages updated
accordingly to get the event and wait on it to preserve the behavior
for now.

[coveralls]

Coverage decreased (-0.001%) to 66.219% when pulling 583fbf7 on feature/DPCTLQueue_Memcpy_async into 832350d on master.

[PokhodenkoSA]

LGTM

[adarshyoga]

Does this change need an update/addition to test cases?

[oleksandr-pavlyk]

Does this change need an update/addition to test cases?

The only change that was needed is in dpctl-capi/tests and was made in this PR. The Python semantics did not change, Python calls remain synchronous. We just replaced previous calls DPCTLQueue_Memcpy(QRef, dst_ptr, src_ptr, num_bytes) with

   ERef = DPCTLQueue_Memcpy(QRef, dst_ptr, src_ptr, num_bytes);
   DPCTLEvent_Wait(ERef);
   DPCTLEvent_Delete(ERef);

[oleksandr-pavlyk]

@reazulhoque Is this change ok to merge?

[reazulhoque]

Not yet. I am still getting segfaults when testing Numba-dppy with dpnp.

[PokhodenkoSA]

Not yet. I am still getting segfaults when testing Numba-dppy with dpnp.

@reazulhoque What versions do you use?

[reazulhoque]

I am using Numba-dppy master and testing locally. This is a matter of updating dpnp glue implementations in Numba-dppy to wait for the memcpy event.

[reazulhoque]

Locally all tests passed. Created PR in Numba-dppy IntelPython/numba-dpex#529