Fix mixture entropy/gibbs_energy single-multi scaling mismatches

[djkees]

Summary

Fixes three related numerical bugs in the single-temperature vs multi-temperature overloads of Mixture%calc_entropy and Mixture%calc_gibbs_energy (source/mixture.f90), so that calling either overload with equivalent physical inputs now gives the same, physically correct result.

djkees#66

Changes

• mixture_calc_entropy_multi divided its result by an extra 1.d3 that mixture_calc_entropy_single did not (the /1.d3 in _single was already commented out). Removed the stray division so both overloads scale identically.
• mixture_calc_gibbs_energy_multi's nj was never normalized by sum(weights), unlike every other _multi routine (calc_entropy_multi, calc_enthalpy_multi) and unlike calc_gibbs_energy_single (built from calc_enthalpy/calc_entropy, both of which do normalize). Added the missing /total_weight normalization.
• A third bug found while verifying the above two fixes against the RP-1311 Example 14 reference data already present (but disabled) in mixture_test.pf: both calc_entropy overloads compute log(pressure/...) using the raw pressure argument directly, without dividing by std_pressure (the 1 bar reference pressure) first. calc_gibbs_energy_multi already does this correctly (pr = pressures(j)/std_pressure), but neither calc_entropy overload did. Without this fix, both overloads agree with each other but disagree with the correct physical value. Added /std_pressure to the pressure term in both overloads.
• Re-enabled the two long-disabled @assertRelativelyEqual assertions in mixture_test.pf's test_calc_properties (calc_gibbs_energy and calc_entropy, both the single- and multi-temperature call forms) — they've been commented out since the initial public release, and now pass with the fix.

Testing

• Not run via ctest/pFUnit locally: pFUnit's own CMake (vendored via GFE) unconditionally excludes RegexFilter.F90 on Windows (if(NOT WIN32) in pFUnit/src/funit/core/CMakeLists.txt) while FUnit.F90's main driver unconditionally does use pf_RegexFilter, so pFUnit itself fails to build on this Windows/MinGW dev environment regardless of this change. CI's pfunit-tests job builds pFUnit on Ubuntu and should exercise mixture_test.pf and equilibrium_test.pf (including test_sv_derivatives, which uses the now-corrected calc_entropy to build an initial-guess constraint value for an entropy/volume-constrained solve) — please confirm CI is green before merging, especially that test.
• Verified numerically outside of pFUnit: built cea_core from this branch and linked a standalone Fortran driver against libcea_core.a that replicates test_calc_properties' exact inputs (RP-1311 Example Case 14 mixture: H2O/O2/H2O(L), T=304K, P=5066.25 Pa). Both calc_entropy overloads now produce 9463.0131180265817 and both calc_gibbs_energy overloads now produce -14830015.9..., matching the reference values 9.46301312d+03 and -1.48300159E+07 in mixture_test.pf to 10 significant figures, and matching each other (single vs. multi) exactly to solver tolerance.

Compatibility / Numerical behavior

• No expected changes to numerical results
• Expected changes (explain and provide validation)

This intentionally changes the numerical output of Mixture%calc_entropy and Mixture%calc_gibbs_energy (both single- and multi-temperature overloads). Per the audit trail above: (1) neither the _multi overloads (which were never exercised by any real caller — see issue #66) nor calc_entropy_single/calc_gibbs_energy_single (used internally by calc_gibbs_energy_single itself, but not otherwise called in production code outside of this module and one test) had validated numerical behavior to preserve; the only reference data available (mixture_test.pf's RP-1311-sourced expected values) was disabled specifically because these functions didn't produce correct output. (2) The one live caller of calc_entropy outside this module, equilibrium_test.pf's test_sv_derivatives, uses it only to construct an initial-guess entropy value fed into a self-consistent constant-entropy/volume equilibrium solve, not to check against a hardcoded expected value — this should remain valid under the corrected calculation, but I could not confirm convergence locally (see Testing above).

---

Drafted with Claude's assistance

• The three bugs and their fixes were derived by direct read of source/mixture.f90:833-967, cross-checked against the normalization pattern used consistently by every other _multi/_single pair in the same file, and against calc_gibbs_energy_multi's existing (correct) std_pressure handling.
• Independently verified numerically (not taken on faith) by building cea_core from this branch and linking a standalone driver program against it that reproduces mixture_test.pf's test_calc_properties inputs exactly; results matched the RP-1311 reference values to 10 significant figures for both the single- and multi-temperature overloads, both before/after comparison for each fix and cross-checked against a hand-derived formula (S = (H-G)/T using the already-correct calc_gibbs_energy_multi).
• Was not able to run the repository's own pFUnit test suite locally due to an unrelated, pre-existing pFUnit-on-Windows build issue (see Testing above) — CI should be checked before merging.

Update

The pFUnit-on-Windows build issue referenced in Testing above (tracked as #163) now has a fix: djkees#167. With that fix, cea_core_test was built and run locally on this branch for the first time: ctest -R cea_core_test -V — 124/124 tests passed, including mixture_test_suite.test_calc_properties (the re-enabled calc_entropy/calc_gibbs_energy assertions this PR's fix targets) and equilibrium_test_suite.test_sv_derivatives (the one live caller of calc_entropy outside this module, flagged above as unconfirmed) — both green.

Also worth noting: this fork's GitHub Actions checks don't appear to run at all (gh pr checks reports nothing on this PR despite .github/workflows/basic_build.yml existing), so the local run above is currently the only automated test evidence for this PR beyond manual review.