Peritheos 0.12.0

This release expands the pressure-scale catalog and its primary-data audits,
adds explicit pressure-coordinate provenance and recalculation, and documents
thermal-pressure conventions in detail. The library contains 650 source EOS
records in 227 material documents; 647 records across 224 materials are
executable, including seven qualified implementations requiring explicit
selection. Scientific source validation, equation reconstruction, and
reproduction of an author's fit remain separate assessments.

Added

• Add dataset pressure-column calibration provenance and per-EOS pressure
  reductions in Python and Rust, with the metadata-driven
  resolve_dataset_pressure API. Migrate the six Dewaele (2004) metal tables
  and their 22 direct 2004/2019 reductions while preserving source observations.
• Add Miozzi (2020) hcp-Fe BM3, Vinet and BM3–MGD source records and all 131
  supplementary observations. Preserve their not_reproduced status and
  explicit-selection access. Add separate nondefault Peritheos thermal refits
  on Tange (2009) Fit3-Vinet and Speziale (2001) variable-q MgO scales, with
  derived inputs, conditional parameter errors, covariance, and sensitivity
  diagnostics. Preserve the author-reported n=2 molar-volume convention in
  the published thermal record; the separate Speziale refit uses n=1.
• Preserve Morard (2026) FeS coefficients, the experimental FeS-VI Pnma
  structure, all 146 supplied thermal observations, and the selected 167-row
  input with 21 literature cold points. Exclude 11 unsuitable quenched rows.
  Add source hashes, EosFit comparisons, and reproducible fitting audits;
  retain the deferred source status and unresolved published ±3 GPa envelope.
• Add the optional Wang (1996) CaSiO3 BM3 plus linear thermal-pressure refit
  with 64 selected observations, equal weights, conditional errors and full
  covariance. Preserve the published records and source-weighting limits;
  recover the source table and Shim (2002) digitized observations with an
  explicit reported-pressure diagnostic.
• Add nondefault Walker (2002) B1, Tateno (2019) Holmes-Pt MGD/linear and
  Sokolova-Pt linear, Chidester (2021) Vinet, and Ma (2024) BM3–MGD KCl
  records. Keep Dewaele as the default. Bundle Walker and official Tateno
  data, Ma's final workbook, 11 acoustic and eight recalibrated Walker inputs,
  and separately identified model grids. Preserve weighting, covariance,
  thermal-convention and upstream-recalibration qualifications.
• Add executable KCl source-pressure reductions and independent diagnostics.
  Trace Campbell-Heinz (1991) to Mao (1978) ruby fluorescence and Walker
  (2002) to Birch (1986) NaCl. Recover the Birch calibration and a conditional
  Walker B1 NaCl replay with run-specific reference volumes; retain the
  unresolved original implementation and uncertainty conventions. Correct
  the Tateno Pt-volume pairing permutation and document source precision.
• Add the complete Matsui (2009) Pt Debye pressure model with a bounded
  electronic-pressure table, source benchmarks and independent quadrature.
  Linear interpolation is an explicit numerical choice; the model does not
  supply an electronic caloric potential or reproduce the original fit.
• Add bounded Yokoo (2009) Au/Pt PVT table-output reconstructions as
  nondefault, explicit-selection diagnostics marked not_reproduced.
  Preserve the separate 0 K cold and 300 K fitted branches, pressure residual
  tables and phase annotations. Document missing electronic inputs and
  fitting data; calculated output grids are not experimental observations.
• Add canonical Fei (2007) Pt and NaCl-B2 thermal pressure records with the
  printed variable-exponent Debye law and unchanged published coefficients.
  Pt retains its separate 300 K record and legacy thermal alias. Bundle
  36 digitized NaCl figure positions representing 24 distinct measured states;
  exclude comparison curves and duplicated cold observations from fits.
• Add independent Fei Au/Pt/NaCl/Ne equation, curve and fit checks, Pt
  weighting/reference-state diagnostics, and published Pt-scale comparisons.
  Fei Au and Ne have qualified numerical parameter agreement; the Pt joint
  and NaCl hot-data diagnostics do not reproduce the reported thermal fits.
  Further Fei Pt investigation remains pending an author response.
• Expose the published Maltby (2024) argon EOS only by explicit selection,
  preserving its not_reproduced status. Retain the Chen (2010)
  printed-constant reconstruction as an explicit-selection DO NOT USE
  diagnostic alongside the independent Peritheos refit. Add the provisional,
  nondefault 49-observation Ross (1986) plus Errandonea (2006) argon refit,
  complete selection provenance, and qualified Xiao (2025) refit findings.
• Audit Speziale (2001) MgO's printed variable-q equations against recoverable
  original thermal constraints and later Fei/Hirose pressure reductions. Bundle
  four Svendsen–Ahrens shock-temperature rows with separate measured and
  calculated columns and explicit volume-normalization diagnostics. Preserve
  published Au/Pt coefficients and reported pressure targets; the complete
  original thermal fit remains unreproduced.
• Verify the Speziale variable-q equations independently with an analytic Debye
  temperature integral and adaptive pressure evaluation. Archive published
  calculated-curve checkpoints, graphical discrepancies, and the distinction
  between verified equations and an unconfirmed historical implementation.
• Expand the Mie–Grüneisen–Debye equation reference with energy and pressure
  derivations, unit conversion, atom-count normalization, reference-isotherm,
  isentrope and absolute-zero conventions, caloric derivatives, the printed
  variable-exponent limitation, and a worked calculation. Refresh source
  audit indexes and record-by-record reproduction ledgers.

Changed

• Rust material loading supports the Matsui and Yokoo tabulated pressure
  mechanisms with bounded tables and lossless interchange. ThermalModel
  and LoadedEos now implement Clone without Copy because the tables own
  their coordinates. Rust callers that copied these enums must borrow them
  or call .clone() explicitly. Python model signatures are unchanged.

Fixed

• Reconcile catalog counts, source/refit audit coverage, Rust fixtures, and
  installed-wheel validation. Qualify reproduction of registered Peritheos
  refits separately from source-author fit parity.
• Fix Python 3.9 typing and documentation/example formatting, and keep
  descriptive-error assertions compatible with Rust 1.99 Clippy.
• Make the Fei source-figure comparison reproducible from bundled images
  and recorded PDF provenance without requiring local Zotero or temporary files.
• Declare the figure-audit image reader in test dependencies, repair source-data
  documentation links, and refresh provenance and catalog regression checks.
• Read and write audit resources explicitly as UTF-8 across platforms and
  preserve Python source line endings for reproducible audit fingerprints.
• Keep source-column checks exact while allowing documented numerical variation
  in calculated audit columns and optimizer-derived quantities across platforms.
  Correct the two NaCl figure-table fingerprints after CSV line normalization.
• Serialize audit artifact and fingerprint paths consistently with forward
  slashes on Windows, and read the remaining catalog-scan test fixtures as UTF-8.