[rfc] performance improvements

[Nytelife26]

Relevant issues

Extension of #1374.

Brief

While performance has been improved greatly since the last release, it is still lacking in some key areas. A full demo run on my system takes ~150ms internally, with a lot of time being lost to encoding calls as a result of using str (uncovered in #1436).

See reference 1 for a full overview of profiling data.

References

1. 

Solution items

• Defer version imports (perf(cli): import version lazily #1438)
• Use a static module register Following further profiling, this makes no difference.
• Read and match against bytes, instead of Python's variable-encoding strs
• Consider using PCRE2 for patterns requiring lookarounds and backreferences