Failing `root_group` test when more than 1 work-group is launched.

[GeorgeWeb]

Describe the bug

The root_group test fails when more than 1 work-group is launched because at the moment the sycl::group_barrier(root_group) is not working correctly (synchronising cross-work-group) for Cuda. The work in-progress, so during that time the root_group will be XFAILed.

Note: Re-enable the test once the correct barrier implementation is in place.

To reproduce

To reproduce, either increase the number of work-groups in the test or use the adapter changes in oneapi-src/unified-runtime#1796 that will return more than 1 work-group from the max_num_work_group_sync kernel-queue-specific info query.

Environment

• OS: Linux
• Target device and vendor: Nvidia GPU
• DPC++ version: latest

Additional context

This issue is used mostly for tracking, so we do not forget to re-enable the test.

[keyRadical]

This can be closed BTW. Here is the PR implementing the root_group barrier and re-enabling the test: #14828