Stabilize Fisher covariance postprocessing

[fujikix1102]

Summary

• convert Fisher matrices with Cholesky factorization and explicit SPD validation
• use the stable covariance helper for 1D uncertainties and ellipse areas
• compute Fisher figure of merit from slogdet
• add tests for reference agreement and invalid matrices

Validation

• pytest -q cosmosis/test/test_postprocessing_statistics.py --disable-warnings --maxfail=1 (3 passed)
• Python compilation and git diff --check passed

[fujikix1102]

Added follow-up commit ee26785: covariance standard deviations now reject non-finite/negative diagonals, ellipse areas reject non-positive/non-finite determinants, and Fisher FOM validates a positive parameter count. Added 3 regression tests; all pass locally.