Peritheos 0.17.0

Added

• Add BM2Reference in Python, Rust, native fitting and EOSMAT, with explicit
  nonzero reference pressure, reference volume and reference bulk modulus.
  Include a complete standalone example and reference-state, energy,
  inversion, fitting and compatibility checks.
• Add the joint coesite-IV/V experimental EOS of Bykova (2018), preserving
  its 32.7 GPa reference and all 16 published observations. Distinguish
  published-curve checks from observation scatter and the unresolved author
  regression; this empirical joint-phase curve is not a default pressure calibrant.
• Document additional Bykova (2018) EOS checks and the diagnostic reconstruction
  of the Dewaele (2008) multi-metal ruby calibration. Preserve published
  coefficients and explicitly retain unresolved source and workflow limitations.

Fixed

• Make the new Bykova audit replays portable across supported numerical
  libraries: bound calculated digitization and diagnostic roundoff while keeping
  source coefficients, pixel readings, observations, hashes and findings exact.
• Identify the coesite-IV Bykova (2018) static entry as a conditional Peritheos
  refit. Separate refit repeatability, source-author reproduction and graphical
  checks; preserve all original EOS coefficients and source observations.
• Retain printed pressure and volume errors in the Fedotenko (2020) Pd audit
  with an explicit ambient-datum treatment. Record conditional coefficient parity
  separately from poor data agreement in the fixed-Kprime=4 fit, which remains
  available for diagnostic selection and is not recommended for quantitative use.

Compatibility and scientific limits

• BM2Reference retains its nonzero reference state in native v4 and explicit
  v3 exports. Readers without this model must reject it; substituting BM2
  changes the pressure law. Existing BM2 calculations remain unchanged.
• This release includes the unreleased 0.16.2 Dewaele provenance and dataset
  corrections listed below. Software checks do not establish physical validation
  or reproduction of an original author's complete regression workflow.