Update llama cmake for custom ops (#10176)

Fix https://github.com/pytorch/executorch/issues/10166

target_link_options_shared_lib has different behavior for macos and
other platforms; seems like whole-archive was causing issues.

Test Plan
`sh .ci/scripts/test_llama_torchao_lowbit.sh`

https://github.com/pytorch/executorch/blob/main/examples/models/llama/README.md#running-with-low-bit-kernels