Better support for unicode numbers

[SethMMorton]

Describe the feature or enhancement

There should be some way to determine all the unicode numbers at start-up with zero runtime cost that does not require releasing a new natsort version.

Provide a concrete example of how the feature or enhancement will improve natsort

Each time a new Python version is released, natsort needs to make an update so that unit tests do not fail. Often, users spot this before I do (#104, #119). The reason I did it this way is because determining this at import time introduces a noticeable slowdown that I am unwilling to make my users incur.

I am looking for ideas/help on how to solve this so that I do not hurt user experience while simultaneously lowering maintenance burden.

[SethMMorton]

I am "fixing" this by making it so that the tests do not fail if this occurs. Instead the tests issue a warning, with instructions on how to fix.