fix: Skip llama_cpp tests on CPUs without AVX2 support

[dsfaccini]

Summary

• Adds AVX2 CPU support detection before running llama_cpp tests
• Gracefully skips tests on machines without AVX2 instead of crashing with Fatal Python error: Illegal instruction
• Fixes flaky CI failures on GitHub Actions runners that lack AVX2 support

Background

The llama_cpp tests have been regularly failing in CI with:

Fatal Python error: Illegal instruction

This happens because llama_cpp uses AVX2 CPU instructions that aren't available on all GitHub Actions runners. Since runners are heterogeneous, the tests are flaky - they pass on some runners and crash on others.

Example failure: https://github.com/pydantic/pydantic-ai/actions/runs/19935894476

Solution

Add a check for AVX2 support on Linux by reading /proc/cpuinfo. If AVX2 is not available, the tests are skipped gracefully rather than crashing.

Test plan

• Pre-commit checks pass
• CI should now skip llama_cpp tests on runners without AVX2 instead of crashing

🤖 Generated with Claude Code

[dsfaccini]

obviously isn't fixing the error, this outlines test is too flaky on CI, some runners obviously miss an instruction that test_outlines triggers but I can't pinpoint which

[dsfaccini]

@DouweM smart code fixes didn't work. To disable it for now you'd need to change the ci.yml

just changing the var:

      # RUN_LLAMA_CPP_TESTS: ${{ matrix.python-version == '3.13' && matrix.install.name == 'all-extras' }}
      RUN_LLAMA_CPP_TESTS: false