util/hybrj: NaN/Inf handling in enorm can cause false convergence

[zingale]

Summary

enorm() returns 0.0 when any input component is NaN or Inf. In hybrj, fnorm == 0.0 is treated as successful convergence (info = 1). This can incorrectly classify invalid numerical states as converged solves.
 

Affected code

• util/hybrj/hybrj_enorm.H: early return 0.0_rt for isnan/isinf
• util/hybrj/hybrj.H: convergence check uses fnorm == 0.0_rt
   

Why this is a problem

A true zero norm and an invalid norm are semantically different. Folding both into 0.0 hides instability and can produce misleading solver status.
 

Expected behavior

Invalid numerical values should lead to a non-converged / invalid-state path, not a successful-convergence code path.
 

Suggested fix

• Return NaN from enorm() for invalid inputs, or
• Add an explicit validity flag / status propagation, and
• Guard convergence checks against NaN/Inf before accepting fnorm == 0.
   

Acceptance criteria

• Invalid residual entries no longer produce info = 1.
• Solver reports a deterministic failure/invalid status for NaN/Inf paths.