Floating point minimum and maximum standard

[ajz34]

Careful implementation of floating point (f32, f64) minimum and maximum should be considered and implemented in a later time.

Current implementation of functions min_all and max_all seems not adhering Array API Standard on NaN values. Also, numpy distinguishes NaN value handling for argmin and nanargmin.

To achieve this, we may either use unstable feature rust-lang/rust#91079 (also c.f. rust-num/num-traits#323), or either implement in our own.

In this current status, we may either using PartialOrd or MinMaxAPI for correct handling of non-nan cases. Using max/min utilities in RSTSR with NaN should be considered as unspecified behaviors.