Reapply "[CI][SYCL] Enable E2E testing for New Offloading Model" (#20634)"

[maksimsab]

Reapply of #20634

This patch has the following changes:

• --param enable_new_offload_model is added to SYCL E2E LIT test configuration
• NewOffloadModel mode run is added to sycl-linux-precommit and sycl-windows-precommit.
• Failing tests in NewOffloadModel mode are marked with xfail/unsupported marker.

[maksimsab]

In this patch sycl-nightly path is extracted in order to not hold the merging of the main part.
XFAIL marker is added to sycl/test-e2e/OptionalKernelFeatures/is_compatible/is_compatible_amdgcn.cpp test because it started failing.

[sarnex]

Do you mind doing a manual run of this?
To do that, push your branch to intel/llvm but not to the sycl branch, to some new branch like nightly_e2e_test or something.

Then, go here
https://github.com/intel/llvm/actions/workflows/sycl-nightly.yml
and on the top rihgt click "Run workflow' and select the branch you made.

After you submit it, hit refresh and the pending nightly job in the list will be the one you just made.

When it's done, please post the log link here.

[maksimsab]

Isn't the testing done in this PR's CI not sufficient? sycl-linux-precommit reuses sycl-linux-build.

[maksimsab]

It appears that sycl-linux-precommit doesn't use sycl-linux-build. So we really should check it manually as you suggest.

[sarnex]

Ah sorry, for some reason I thought this change was for the nightly only.

If we look at the CI results in this PR
https://github.com/intel/llvm/actions/runs/21002533524/job/60379060956?pr=21036
https://github.com/intel/llvm/actions/runs/21002533381/job/60379088909?pr=21036

We can see it seems to be running the new offload model tests, so I think it's working

sycl-linux-precommit should use sycl-linux-build here https://github.com/intel/llvm/blob/sycl/.github/workflows/sycl-linux-precommit.yml#L56.

[maksimsab]

@sarnex

Here is the result for linux: https://github.com/intel/llvm/actions/runs/21066409654/job/60584763744
Failing tests are identical for NewOffloadModel and OldOffloadModel.

Here is the result for windows: https://github.com/intel/llvm/actions/runs/21066409654/job/60584763801

I have found one mistake which will be fixed soon.

[sarnex]

lgtm, thx!

[sarnex]

lol i am in too many code review groups