Remove llava from ci_expected_accuracy as it's flaky (#121322)
https://github.com/pytorch/pytorch/pull/121029 added it into the CI but the test is flaky on hud. It alternates between fail_accuracy and fail_to_run
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121322
Approved by: https://github.com/desertfire
