[Divo-Lee]

Fitting with --audit-ceiling 400 previously wrote validation_status: audited-e0-le-100, and both empirical parameter loaders only accepted that fixed label. The fitter now records the actual ceiling, and experimental/predicted loaders share validation of the audit metadata and fitted range.

Existing fixed-100 labels remain compatible with an explicitly recorded larger audit ceiling, so previously fitted NuFold parameters do not need refitting for this metadata fix. Other label/ceiling mismatches, non-finite ceilings, and fitted ranges beyond the audited range are rejected. Scoring, Gumbel fitting, empirical mapping, and extrapolation are unchanged; no database-specific parameters are added.

Validation:

• 37 targeted tests passed from source, covering ceilings 100/400, legacy compatibility, invalid metadata, fitting, attaching parameters, and search.
• Built and installed the wheel into an isolated directory; the same 37 tests passed against the installed package.
• Full Windows suite: 117 passed, 2 skipped, 1 failed. The existing legacy calibrate CLI test launches a Unix shebang-only mock US-align executable, which Windows rejects with WinError 193. That test and execution path are unchanged by this PR.
• git diff --check passed.

This PR fixes the metadata issue only; it does not implement same-cluster audit prefetch or change audit convergence limits.