added LAST_VALID_SOLUTION convergence enum

[markleader]

Summary

Add a new soft-failure status, CEA_LAST_VALID_SOLUTION, for the incident-equilibrium shock recovery path when CEA retains a last valid incident state without converging the shock iteration. Closes #76.

Motivation:

• Preserve the meaning of solution.converged == false for strict shock non-convergence
• Make the C/Python API distinguish between a hard non-converged failure and a retained last-valid shock state
• Avoid incorrectly signaling full success for fallback shock states

Impacted components:

• Fortran shock solver internals
• C binding error/status mapping
• Python binding warning/error handling
• C/Python docs and regression tests

Changes

• Added shared error code CEA_LAST_VALID_SOLUTION to the C API enum surface
• Added a private internal shock solve status on ShockSolution to distinguish:
  • converged shock solve
  • hard non-converged failure
  • retained last-valid incident-equilibrium state
• Kept ShockSolution%converged semantics unchanged:
  • true still means the shock iteration converged
  • retained last-valid states still report false
• Updated cea_shock_solver_solve to return:
  • CEA_SUCCESS for converged shock solves
  • CEA_LAST_VALID_SOLUTION for retained incident last-valid states
  • CEA_NOT_CONVERGED for other non-converged shock outcomes
• Updated the Python binding to:
  • expose LAST_VALID_SOLUTION
  • emit a RuntimeWarning for this soft-failure code
  • keep soln.last_error interpretable without changing soln.converged
• Added comments/docs clarifying that a last-valid retained state is not a converged shock point
• Added regressions for:
  • internal shock status behavior
  • direct C API return-code behavior
  • Python warning + last_error behavior

Testing

• Ran ctest -R cea_core_test -V
• Ran Python binding tests against the freshly built extension from build-dev:
  • source/bind/python/tests -> 51 passed

Compatibility / Numerical behavior

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

Expected behavior changes:

• C shock callers can now distinguish retained last-valid incident states from hard CEA_NOT_CONVERGED failures.
• Python now warns with LAST_VALID_SOLUTION for this specific soft-failure path while keeping solution.converged == false.
• For retained last-valid incident states, the incident pressure array entry is now restored consistently with the retained shock ratios instead of being left stale/zero.

Validation:

• Added a direct C regression covering CEA_LAST_VALID_SOLUTION
• Added a Python regression asserting warning text, last_error, and converged == false
• Added a Fortran regression asserting the internal retained-state status remains non-converged