Add a fast_split module

[ArthurZucker]

This removes dependencies on Unicode, Regex, Regex-Automata by unrolling.
On 10K different models only a handful of Rules are actually needed.
Each Pattern is a combination of different Rules. Rules use Matchers to get info on the bytes (is it a letter/digit).

FastLetter vs Letter  (15200 bytes)
  bitmap  : 0.583 ns/byte (  1714 MB/s)
  unicode : 6.293 ns/byte (   159 MB/s)  speedup 10.79x

FastNumber vs Digit  (4000 bytes)
  bitmap  : 0.400 ns/byte (  2503 MB/s)
  unicode : 5.410 ns/byte (   185 MB/s)  speedup 13.54x