refactor: introduce MathUtilModule, move is_same, add tests

[wpbonelli]

The GenericUtilitiesModule currently has a function is_same() which uses a symmetric relative comparison for floating point equality. Move it to new MathUtilModule as GenericUtilitiesModule is mostly simulation related and depends on SimVariablesModule etc. Modify floating point equality routine and rename to is_close().

Motivation

Relative comparisons get more sensitive to difference near zero holding epsilon equal, and if either value is 0, equality is only reported with epsilon >= 1. This can cause difficulties if it's unknown whether values to compare will be close to 0.

A common approach is to fall back to an absolute comparison like Python's math.isclose() and numpy.isclose(). Introduce function is_close() doing this.

Python stdlib math and numpy use different abstol defaults: math.isclose() sets the default absolute tolerance to zero, motivation that there is no universally reasonable default. The new is_close() here follows this but I'm not sure that's the best way.

An option is also provided to toggle asymmetric comparison, e.g. with a known reference value to which relative difference should be scaled

Places (near-)zero comparisons are or might be made:

• utils/mf5to6/src/Preproc/ObsBlock.f90 checks explicitly against zero
• TimeSeries and TimeArraySeries time comparisons
• coordinate comparisons in GridSorting.f90
• point_in_polygon() could support fuzzy boundaries

Reference

• https://peps.python.org/pep-0485/
• https://stackoverflow.com/a/32334103/6514033
• https://numpy.org/doc/stable/reference/generated/numpy.isclose.html
• https://github.com/fortran-lang/stdlib/blob/master/src/stdlib_math_is_close.fypp

[langevin-usgs]

@wpbonelli, this looks good to me. If we switch to is_close() do we still need is_same()? And I'm sure it's fine, but maybe in TestMathUtil we should explicitly verify that comparison with DNODATA works as this is used in csub?

[wpbonelli]

Removed the older is_same() as it is no longer needed. Since the default abstol is zero, this should behave identically. It may be worth revisiting some uses and considering whether to use a nonzero abstol. Also added some test cases comparing with DNODATA as used in gwflak, I only saw one double precision comparison (with 0) in csub.