Bug/deton unconverged

[markleader]

Summary

Fix Issue #42 by resetting transient equilibrium Newton state on each EqSolver_solve call so reusing an EqSolution does not carry stale iteration history into the next solve.

This addresses history-dependent non-convergence observed in:

• HP → TP reuse flow (O2/C2H4 case)
• Reused TP sequences with condensed phases (Air/C2H4/AL case)

In addition, this PR now hardens the non-converged recovery path so failed/singular solves do not poison subsequent solves:

• EqSolution now stores a last-stable warm-start snapshot
• If the previous solve was non-converged, the next solve restores that stable snapshot before iterating
• Expected non-converged solver paths return converged=.false. (instead of abort) so sweeps can continue

Changes

• Core solver robustness:

  • Added EqSolution_reset_iteration_state(soln) in source/equilibrium.f90
  • Called it at the start of EqSolver_solve (after constraints setup, before matrix/phase iteration)

• Reset scope is intentionally surgical (transient fields only):

  • dln_nj(:), dnj_c(:), dln_n, dln_T, dpi_e
  • convergence flags (gas_converged, condensed_converged, moles_converged, element_converged, temperature_converged, entropy_converged, pi_converged, ions_converged, converged)
  • does not reset warm-start state (T, nj, ln_nj, active condensed set, pi, pi_prev)

• Additional core recovery/state lifecycle updates in source/equilibrium.f90:

  • Added last-stable warm-start snapshot fields on EqSolution:
    • T_seed, n_seed
    • nj_seed(:), ln_nj_seed(:)
    • is_active_seed(:), active_rank_seed(:)
    • j_sol_seed, j_liq_seed, j_switch_seed, last_cond_idx_seed
  • Added helper routines:
    • EqSolution_save_seed(soln)
    • EqSolution_restore_seed(soln)
  • Initialized snapshot state in EqSolution_init
  • Updated EqSolver_solve entry logic:
    • if prior solve did not converge, restore last stable snapshot before starting new solve
  • Snapshot updates are success-only:
    • save seed only when solve exits converged
  • Normalized expected non-converged exits in EqSolver_solve:
    • too many singular matrices
    • failure to establish condensed species set
    • maximum iterations reached without convergence
    • now return with soln%converged=.false. (and restore reduced elements) rather than aborting

• New core regressions in source/equilibrium_test.pf:

  • test_issue42_reuse_solution_hp_then_tp
  • test_issue42_reuse_solution_condensed_tp_grid
  • test_reuse_solution_recovers_after_nonconverged_point

• New Python regressions in source/bind/python/tests/test_issue42_solver_history.py:

  • test_issue42_detonation_no_inner_eq_failure
  • test_issue42_reused_eqsolution_matches_fresh

Testing

• Ran:
  • ctest -R cea_core_test -V
  • python test_main.py
  • python -m pytest
  • .py files attached to the issue

Compatibility / Numerical behavior

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

Expected behavior change is limited to solver robustness when reusing EqSolution objects across solves:

• Previously history-dependent non-convergence cases now converge in core regression tests.
• Reuse-vs-fresh solutions agree within tight tolerances in new tests.
• Non-converged expected sweep points now return converged=.false. instead of aborting the process.
• Subsequent solves after a failed point are seeded from the last stable converged snapshot, not the failed iterate.
• No API/signature changes in Fortran/C/Python interfaces.