[L0] Implement Support for zeInitDrivers

[nrspruit]

• As of v1.10 of the L0 spec, zeInit and zeDriverGet is the old init pathway and the desired init api is zeInitDrivers. This new api allows for multi heterogenous drivers to coexist in a single L0 Process.

[nrspruit]

@nrspruit This is marked ready to merge but two L0 jobs have failed (not counting the E2E tests) - are these failures expected?

@callumfare the errors are unrelated and fail sporadically in multiple PRs.

I can rebase and repush the changes to see if I can clear it out. I will remove the read to merge until that is complete.

[nrspruit]

@winstonzhang-intel , I found a couple of things that I needed to change, can you re-review?

[nrspruit]

@callumfare , I will need @pbalcer's help to determine why only the static lib no platform test is hanging when it passes everywhere else.

[nrspruit]

@callumfare , I will need @pbalcer's help to determine why only the static lib no platform test is hanging when it passes everywhere else.

Based off an offline discussion, because the loader test error for loader_platform_test_no_platforms cannot be reproduced elsewhere we are disabling it until the issues with ctest can be worked out.

[nrspruit]

There was a race condition in the code, I have re-enabled the test that was failing on windows to see if that change fixes the issue. If so, then I will add back in the ready_to_merge, awaiting update of the CI results to be sure.

[nrspruit]

There was a race condition in the code, I have re-enabled the test that was failing on windows to see if that change fixes the issue. If so, then I will add back in the ready_to_merge, awaiting update of the CI results to be sure.

the fix to the code did not address the race condition with the loader no platforms test, I disabled the test again. Once the CI finishes I will add the ready_to_merge label again since this has been reviewed again offline.

[nrspruit]

Based off the intel/llvm CI result here: intel/llvm#16034 and the updated results in the CI, this is ready to merge again.

The failures in the e2e CI in UR are unrelated and have been seen before.