[regexp] Split code point set builders and match start analysis out of compiler

[Hans-Halverson]

Summary

Refactor the RegExp compiler to focus on just bytecode generation:

• Match start analysis moved alongise rest of match start code, encapsulated in a new MatchStartAnalyzer
• Make RegExpFlagsStack its own utility struct used in both analysis and compilation traversals
• Move the logic for converting a code point or character class to full code point set (aware of case sensitivity and unicode modes) into a new module under CodePointSetBuilder, only exposing utilities to create the full set for a code point or class.
• Refactor some of the builder methods to pass down CodePointInversionListBuilders while possible

Tests

• CI