test: query_clarification times out on CPU during full test run

[planetf1]

Summary

test_run_transformers[query_clarification] times out after 180s during torch.nn.Linear.forward (matrix multiplication) on CPU/MPS without GPU.

Context

Discovered during PR #1294 (aLoRA input_ids fix) while verifying the full test suite. This is a separate pre-existing issue that also exists on upstream/main without the fix.

Details

• Model: ibm-granite/granite-4.1-3b
• Context: 4 conversation turns + 12 retrieved documents (~2000+ tokens)
• Generation: max_completion_tokens: 512
• Hardware: Apple Silicon (MPS), no CUDA

The test gets stuck in F.linear() doing a forward pass on a ~2000-token context. This is raw compute, not a code bug.

Why it matters

The test is already gated on CI (gh_run == 1 xfails it), but it blocks local full-suite runs and any developer running pytest -m "not qualitative" without realizing this one test will take 3+ minutes on CPU.

Proposed fixes

1. Add @pytest.mark.slow to exclude it from the fast loop
2. Add a CPU-only skip: pytest.skip("query_clarification takes >180s on CPU")
3. Increase the timeout for this specific test