Java Regex Refactor

[lmasroca]

Refactored Java regex character classes in both grammar and visitor:

• Used new lexer mode (CHAR_CLASS_MODE) to avoid having to manually enumerate all the default lexer tokens allowed in classAtomNoDash in the Java parser. This simplifies that parser rule and allows us to remove code in visitNonemptyClassRanges that was breaking those default tokens back down into literal characters. This also allows us to remove && as a default mode lexer token, making it only a token within character classes.
• Moved \p{...} escapes from CharacterEscape (which represents one character) into CharacterClassEscape (which represents predefined character classes), since \p{...} escapes represent predefined character classes.
• Made classEscape have its own behavior instead of wrapping atomEscape, and extracted shared code into a method used by both. This simplifies visitClassEscape by skipping having to throw on back references since those are now rejected at lexer level (see related note below) as now back references no longer have a token on that context (CHAR_CLASS_MODE).
• Added a test to check that character classes and related features are still working as intended.

Note: Since one of the new tests checks that [\1] is rejected at the lexer level, RegexHandler.ThrowingErrorListener.syntaxError(...) had to be updated. That method declared offendingSymbol: Any, but it is actually null when a lexer error is thrown (as there is no symbol at that point), so we changed it to Any?.