transformer qwen3 Porting to optimum-habana

[rkumar2patel]

This PR ports qwen3 model testing to the optimum-habana library for G2 hardware acceleration. The testing suite provides comprehensive coverage for the qwen3 model implementation on Habana Processing Units (HPUs).

Key changes:

Added complete test suite for qwen3 model with HPU-specific adaptations
Configured HPU device targeting and Habana-specific transformers integration

[regisss]

LGTM.
Will let @karol-brejna-i and @gplutop7 approve too before merging since these tests are not part of the Github CI.

[github-actions]

The code quality check failed, please run make style.

[regisss]

The code quality check failed, please run make style.

@rkumar2patel You'll probably need to merge the latest main branch into yours and then run make style

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[gplutop7]

Suggestion - Add the guards:

if not is_habana_available():
    pytest.skip("HPU not available", allow_module_level=True)

if os.environ.get("HF_HUB_OFFLINE") == "1":
    pytest.skip("Requires HF Hub access (HF_HUB_OFFLINE=1).", allow_module_level=True)

[rkumar2patel]

Since this guard is not present in other optimum-habana models, I have followed the same approach for consistency. If we determine it is required, we should incorporate it across all models rather than selectively, to avoid any inconsistency.