[codex] Fix GX eik input handling and subprocess errors

[rmchurch]

What changed

• force generated GX inputs to use geo_option = "eik" so DESC's plain-text geometry files match the GX input mode
• wrap GX subprocess failures and timeouts in readable runtime errors that include stdout.gx / stderr.gx context
• add targeted tests for input rewriting and subprocess error reporting

Why

The GX wrapper was generating a plain-text geometry file while allowing templates to keep geo_option = "nc". That made GX fail with an opaque nonzero exit, and the callback path surfaced it as a raw subprocess failure instead of a useful GX-specific error.

Impact

This makes the GX integration robust against NC-based templates and preserves actionable failure context when GX exits nonzero.

Validation

• /global/cfs/projectdirs/m4505/rchurchi/conda-envs/desc-gx/bin/python -m pytest tests/test_external_gx.py -q

[YigitElma]

Thanks for this @rmchurch! I was naively expecting the user to put eik, but I guess people use the same template file for multiple cases.

If this is ready, I can approve/merge it to yge/gx, and create a GX wrapper PR later. Does this sound good to you?

[YigitElma]

We actually don't require writing tests for the desc.external module. But these look good to me. We can decide to keep them or not in the actual PR for the yge/gx branch.

[YigitElma]

Hi @rmchurch, if you can resolve the merge conflicts, I can merge this to the other branch and finish it up on the other branch.

[YigitElma]

In the other branch, I switched to using a loop to update the file content. This is mainly because regex operations are very hard to read, and some other developers wanted that.

@rmchurch Can you switch to the other method when resolving the merge conflicts?