[IPEX deprecation] Process to update upstream Pytorch pin and automation around it

[vlad-penkin]

Run validation worklfow for latest commit in main branch, which will include:

• UT + Tutorials validation matrix - Supported Python versions x Driver Streams (Rolling, LTS)
• Inductor for all supported Python versions on Rolling
• E2E Accuracy LTS and Rolling for selected Python version
• Benchmarks for selected Python version

Workflow run frequency

• Monday and Wednesday or ad-hoc

Tracking

• manually via GH issues We'll start from manual check and consider automation later

Baseline comparison with current UT/Tutorials and E2E results

• We'll start from manual check and consider automation later

[pbchekin]

Inductor for all supported Python versions on Rolling

Only one Python is fine. Also inductor tests can be a gateway for other tests: execute it first, then if successful execute rest of the tests.

[kwasd]

Done 3 of 4 (merged to llvm-target), but still need #1905 to schedule benchmarks for selected Python version: this adds pytorch_ref handling to the triton-benchmarks.yml.

[kwasd]

Inductor for all supported Python versions on Rolling

Only one Python is fine. Also inductor tests can be a gateway for other tests: execute it first, then if successful execute rest of the tests.

See #2161

[kwasd]

Example https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/10769983032

[kwasd]

Will continue in #2172