Fix candidate gating on PSPL fit quality

[NunotaKansuke]

Summary

• Gate the final is_candidate flag on the PSPL fit-quality decision.
• Add a regression test for a detected bump with an unacceptable PSPL fit.

Root cause

fit_pspl_candidate already returns is_candidate=False when the reduced PSPL chi-square exceeds good_pspl_chi2, but the pipeline ignored that value and derived the final flag only from the bump scan and vetoes. This allowed poor PSPL fits to be reported as candidates.

Validation

• pytest -q — 16 passed
• ruff check src tests — passed
• The change was merged into NunotaKansuke/aethra main before opening this PR.

The bug was reproduced on a Roman simulation event where chi2_red_pspl=2.979 exceeded the default threshold of 2.5, while the pipeline previously returned is_candidate=True.