Fix for crash reported in gh-1654

[oleksandr-pavlyk]

Closes gh-1654

The reason behind the crash was out of bound access to shared local memory accessor.

This also fixes a crash in dpt.sort on CUDA device for sorting of 256 elements of floating point numbers.

• 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?
• Have you checked performance impact of proposed changes?
• If this PR is a work in progress, are you opening the PR as a draft?

[github-actions]

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

[github-actions]

Array API standard conformance tests for dpctl=0.17.0dev0=py310h15de555_336 ran successfully.
Passed: 888
Failed: 17
Skipped: 91

[coveralls]

coverage: 87.948%. remained the same
when pulling 58dfb5c on fix-for-crash-gh-1654
into 8b42313 on master.

[ndgrigorian]

CUDA tests pass, so LGTM. Nice fix to get in this version.

[oleksandr-pavlyk]

The example that was crashing on CUDA device:

import dpctl.tensor as dpt
x = dpt.arange(256, dtype="i4", device="cuda:gpu")
dpt.sort(x, descending=True) # crash

[antonwolfy]

I checked locally DPNP tests relating to sort and argsort functions and they passed without any issue.
Also the gh-issue reported as #1676 isn't reproduced anymore locally for me with the proposed dpctl fix.

Thank you @oleksandr-pavlyk for resolving the issue!