fix: propagate CDOpt runner failures safely

[christiek673-mAth]

Summary

• Harden the three CDOpt runner generators against symlinked or invalid output paths.
• Validate generated runner parameters before execution.
• Propagate training/solver/verification failures through the generated runner exit status.
• Add explicit execution, solver, verification, and completion fields to generated summaries.
• Add regression tests for generated runner compilation, failure exit propagation, and symlink protection.

Validation

• python3.11 -m unittest discover -s tests -q
• Result: 32 tests passed, 5 skipped because optional numerical dependencies are unavailable locally.
• git diff --check passed.

The repository's existing test workflow installs the OSQP test dependency; the skipped tests are dependency-gated numerical tests, not failures.

[christiek673-mAth]

Hi maintainers, could you please review this PR? It hardens the three CDOpt runner generators against unsafe output paths, validates parameters, and propagates execution/solver/verification failures through the generated runner exit status. Local Python 3.11 validation passed: 32 tests passed, 5 dependency-gated numerical tests skipped because optional OSQP/NumPy dependencies were unavailable locally, and git diff --check passed. The fork PR workflow currently shows action_required with no jobs started; could a maintainer please approve and run the Validate skills workflow?