Fixed inconsistency between the RealField::min_value docs and functionality.

[wdlea]

The docs for the function state "The smallest finite positive value representable using this type.", but f32::MIN & f64::MIN are both the smallest values representable(and are thus negative). The f32::MIN_POSITIVE & f64::MIN_POSITIVE are the constants with the smallest values for their respective types that are still positive.

This could also be an issue with the docs, and the RealField::min_value() function is actually meant to return the smallest value for the type, in which case this pull request would break projects depending on this function.