Peritheos 0.13.0

This release expands thermal EOS coverage for forsterite, diopside, brucite and
rubidium halides, adds Fe-bearing phase E and an optional NaCl-B2 refit, and
improves uncertainty propagation and lossless interchange. It contains 672
stored EOS records in 232 material documents, with 669 executable records
across 229 materials. Published models and independent refits retain their
individual scientific qualifications.

Added

• Add an opt-in NaCl-B2 thermal refit of Fei (2007)'s 24 unique digitized
  observations, with explicit cold-then-hot fitting and conditional statistics.
  Preserve published coefficients and the default; original-author thermal-fit
  parity remains unresolved.
• Add separate brucite curved-expansion refits: four isotherms retain the
  published moduli, while a continuous version uses the published slopes and
  explicit 300 K anchors. Preserve original records/defaults and shared
  expansion covariance. Correct finite-difference uncertainty steps for very
  small coefficients, including temperature-dependent expansivity curvature.
• Add brucite over Fukui (2003)'s measured lattice-average pressure range,
  with a directly inspected F-free neutron structure. Preserve all published
  coefficients and observations. Separate Peritheos unweighted 873 K and
  continuous thermal table-data refits include conditional covariance;
  the original 873 K normalization mismatch and source-slope reconstruction
  remain qualified explicit selections.
• Add the measured Fe-bearing phase E of Crichton and Ross (2000), with the
  published BM3 coefficients, all eleven primary rows and reported errors,
  and a documented phase-specific diffraction peak-position fallback. Record
  fit parity within reported uncertainty, coefficient and curve differences,
  and the reproducible residual plot; keep exact-workflow provenance separate.
• Support a linear temperature derivative of K0_prime in
  ThermalReferenceStateEOS / AlphaKT through optional dKprime_dT (K^-1),
  including native evaluation, fitting, uncertainty propagation and .eosmat
  interchange. Omission retains existing behavior; unsupported reference
  families reject nonzero shifts.
• Add two Katsura (2022) forsterite thermal records, retaining the full-precision
  author-script option and the printed-table alternative with scoped conditional
  numerical parity. Preserve the static default and original-2009 fit hold.
• Add diffraction-ready diopside with independent Thompson (2008) static and
  Zhao (1998) thermal records. Retain original numerical tables and a separate
  graph-corroborated corrected analysis dataset; exact author workflow is unverified.
• Add Farla (2025) B2 RbCl/RbBr/RbI thermal records and primary numerical datasets.
  Retain conditional fit similarity, benchmark precision and composite-CsCl
  calibration qualifications; preserve the existing RbCl static default.
• Integrate the optional combined Hirose/Fei Au refit and conditional Sakai
  (2025) Au central-value replay into the scientific ledgers without claiming
  original-author uncertainty or complete sequential-calibration parity.
• Add a generated per-material coverage overview and active material-gap queue.

Fixed

• Retry stalled RbCl rounding diagnostics at an explicit numerical precision
  floor, requiring genuine solver convergence, feasibility and a consistent
  objective before accepting the result. Source observations are unchanged.
• Correct the siderite uncertainty regression and documentation using independent
  analytic derivatives; small-coefficient finite differences no longer inflate
  the conditional pressure error.
• Preserve archived brucite refit coefficients and run provenance when rebuilding
  catalog records. Bound numerical replay variation across Python/SciPy versions
  while retaining exact source facts and strict runtime equation checks.
• Make brucite regression checks independent of local staging files and include
  their fixtures in the standalone Rust crate. Keep unfinished Brown NaCl and
  Fe-ringwoodite investigations outside this release.
• Preserve UTF-8 source text, archived path syntax and canonical dataset bytes
  across platforms, including regeneration on Windows.
• Reconcile source/refit ledgers and the material manifest for all 672 stored
  records in 232 documents (669 executable records across 229 materials).
• Preserve small floating-point EOS coefficients through Rust JSON round trips
  using the exact-rounding reader; retain source numeric values and defaults.
• Reconcile catalog tests, source links and documentation counts after the first
  material-gap batch.