CI: retry Pkg.test() once on transient PythonCall init failures

[ChrisRackauckas-Claude]

Summary

The OptimizationSciPy (and occasionally OptimizationPyCMA) jobs intermittently fail at the precompile stage with

InitError: UndefRefError: access to undefined reference
 [11] __init__()
    @ PythonCall.JlWrap ~/.julia/packages/PythonCall/.../src/JlWrap/JlWrap.jl:70
during initialization of module JlWrap
in expression starting at .../OptimizationSciPy/src/OptimizationSciPy.jl:2

This happens with the same PythonCall version (v0.9.31) on runs where other jobs succeed — e.g. OptimizationSciPy, lts passes on master run 24144818227 but the same job fails on PR #1169 run 23171032225. It's an upstream PythonCall flakiness around precompile cache deserialization, not a code change in this repository.

Approach

Wrap Pkg.test() so that, on a PythonCall + JlWrap / UndefRefError-shaped exception:

1. The precompile cache directories for PythonCall, OptimizationSciPy, OptimizationPyCMA, and SciMLBasePythonCallExt are removed (they're what carry the stale JlWrap state across runs).
2. Pkg.test() is retried exactly once.

Any other exception is rethrown immediately, so genuine test failures are still surfaced. The retry only fires when the error matches both PythonCall and one of JlWrap / UndefRefError, so it will not mask unrelated test bugs in any group.

Test plan

• Detection function unit-tested locally against the CI error string and against unrelated BoundsError / generic ErrorException (only the PythonCall init shape returns true)
• Embedded Julia script parses (Meta.parseall) cleanly
• YAML validates with yaml.safe_load
• Live CI run

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Closing per maintainer feedback. Retry isn't the right approach — replaced by #1182, which sets JULIA_NUM_PRECOMPILE_TASKS=1 for the python-using groups to serialize precompile workers and remove the underlying PythonCall.JlWrap.__init__ parallel-init race that was the actual source of the failure.