Fix covariance QA and posterior transform plotting

[botjean]

Summary

This PR fixes several issues encountered when running the covariance-enabled Bayesian workflow on current table snapshots:

• ignore filesystem sidecars / stray non-table files during table loading
• add reproducible emcee seeding and persist lightweight chain QA diagnostics
• fix plot_mcmc compatibility with the current MCConfig
• add config-driven posterior transform plotting from unit-space to physical parameters
• make observable panel plotting robust when plot_panel_shapes is omitted
• replace / add regression tests covering these paths

Details

Table loading

• Skip :Zone.Identifier sidecars and other non-table junk files in Data/, Design/, and Prediction/
• Add a loader integration test covering sidecars and stray files

MCMC reproducibility and QA

• Add optional random_seed support to the emcee sampler
• Seed both initial positions and emcee's internal RNG
• Persist random_seed and chain_diagnostics into mcmc.h5
• Add tests for deterministic seeded short chains
• Add tests for chain diagnostics and input validation

Posterior plotting

• Fix acceptance-fraction plotting so it no longer assumes config.n_walkers
• Add plot_posterior_transform.py to generate:
  • unit-space 1D/2D posterior plots
  • physical-space 1D/2D posterior plots
  • linear-axis variants for transformed parameters when useful
  • posterior_unit_and_physical.npz
  • posterior_transform_summary.txt
• Drive the transform from YAML via parameterization.<name>.physical_posterior_transform
• Add guards for invalid / nonpositive values on log-scaled transformed axes

Plot utility robustness

• Fall back to an automatically derived subplot grid when plot_panel_shapes is not specified

Tests

• Replace the broken fixture-dependent outlier smoothing test with a self-contained synthetic test
• Add transform tests, including inverse-step support and log-axis filtering behavior

Validation

• pytest: 16 passed
• Verified real plotting reruns on saved detailed and compact outputs