Improve performance of array metadata transfer to device

[oleksandr-pavlyk]

Change to use USM-host allocation for packed shape/strides on the host.
USM-host allocations take less time to copy them into USM-device allocations.

The difference is going to be especially pronounced on discrete GPU devices.

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?

[github-actions]

View rendered docs @ https://intelpython.github.io/dpctl/pulls/912/index.html

[coveralls]

Coverage remained the same at 82.02% when pulling eef632e on use-usm-host-allocator into 0f0c926 on master.

[github-actions]

Array API standard conformance tests failed to run for dpctl=0.14.0dev0=py310h8c27c75_118.

[diptorupd]

@oleksandr-pavlyk re: "USM-host allocations take less time to copy them into USM-device allocations."

Less time compared to what? previously were we using regular malloc?

[oleksandr-pavlyk]

q.copy<T>(standard_alloc_vector.data(), usm_device_ptr, n) is slower than q.copy<T>(usm_host_alloc_vector.data(), usm_device_ptr, n). The PR changes to use the latter.

[github-actions]

Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞

[github-actions]

Array API standard conformance tests failed to run for dpctl=0.14.0dev0=py310h8c27c75_118.