[Test] UT for linear method patch

[rjg-lyh]

What this PR does / why we need it?

UT for linear method patch.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

None.

[MengqingCao]

This relys on multi-npu ci machine, cc @Potabk
@rjg-lyh Please firstly make CI happy with code format check, thx!

[Potabk]

This relys on multi-npu ci machine, cc @Potabk @rjg-lyh Please firstly make CI happy with code format check, thx!

Currently, all our unit tests are run on a single NPU. and we now have multiple NPU runners but cannot flexibly allocate them according to test requirements. here is a solution: mark a test as single or multiple to adapt to the runner's filtering， just like the example following:

@pytest.mark.multi
@pytest.mark.parametrize(
    "model, distributed_executor_backend", [
        ("Qwen/QwQ-32B", "mp"),
    ])
def test_models_distributed(
    vllm_runner,
    example_prompts,
    model: str,
    distributed_executor_backend: str,
) -> None:
    dtype = "half"
    max_tokens = 5
    with vllm_runner(model,
                     dtype=dtype,
                     tensor_parallel_size=2,
                     distributed_executor_backend=distributed_executor_backend
                     ) as vllm_model:
        vllm_model.generate_greedy(example_prompts, max_tokens)

and then we will run the multi-npu test case with command pytest -sv -m multi test/