verify float computations and explain output differences

[sarnold]

Verify IEEE math compliance and explain output grid differences from older g77 builds. Current output has occasional differences in one value per row off by one least digit compared to sample grids (less than %10 of the rows).

• Fix floating-point exception:
  Note: The following floating-point exceptions are signalling: IEEE_INVALID_FLAG
• Verify float comparisons and convergence (legacy F77 code does not use newer math intrinsics such as EPSILON)
• Explain differences and decide which is correctly computed (sample grids or new output)