[JVM] Improve handling of unicode numbers

[usev6]

[usev6]

I didn't see any additional spectest breakage from this changes.

[usev6]

I'm tempted to merge this even without getting another pair of eyes.

Sure, the addition to unipropcode and getuniprop_str look quite ad hoc and are nothing in comparison to what MoarVM does there. (Even though I picked the same numbers as MoarVM for UNIPROP_NUMERIC_VALUE_NUMERATOR and UNIPROP_NUMERIC_VALUE_DENOMINATOR.)

But on the other hand the current situation (described in https://github.com/rakudo/rakudo/issues/4760) is clearly LTA. The usage of nqp::getuniprop_str in Rakudo's method numish (in https://github.com/rakudo/rakudo/blob/30a9d3a40a/src/Perl6/Actions.nqp#L8526-L8527) is quite central -- and the JVM backend can do better there. I'd hope the small patch wouldn't hinder any future improvements in that area.

[usev6]

@lizmat thanks for looking!