[BUG] RegExp parse fails to parse character ranges containing escaped characters

[andygrove]

Describe the bug
The regular expression [\a-z] does not get parsed correctly because we currently only support character ranges with non-escaped characters. The transpiler then converts this to [\a\-z] when passing it to cuDF, resulting in incorrect results.

Steps/Code to reproduce bug
This was found by the new test code in #4504

Expected behavior
This expression should produce the same results on CPU and GPU.

Additional context
None

[andygrove]

Note that I don't think this is a very common use case so not the highest priority but needs fixing to unblock the new AST-based fuzzing approach.

Some notes on how I would implement this, in case someone else picks this issue up:

• RegexChar and RegexEscapedChar should be combined as RegChar(ch: Char, escaped: Boolean)
• RegexCharacterRange should be updated to accept RegexChar instead of Char
• Add unit tests to RegularExpressionParserSuite for parsing [\a-z] and assert that this produces a character range
• Update RegexParser as required

[anthony-chang]

Related to this issue: with the newly added support of predefined character classes (eg. \s, \h, \v etc), we need to handle the cases where these are nested in another character class (eg. [abc\s]).

cuDF doesn't support nested character classes but we should be able to expand and flatten the nested predefined classes, ie. [abc\s] becomes [abc \t\n\x0B\f\r]