Release v5.2.0

[RomeCore]

New features:

• Add Scan() methods returning interleaved sequence of raw text
  segments and parsed matches via Or<StringSegment, ParsedRuleResultBase>
• Add SplitSegments() methods to split input by parser rule matches
  (analogous to Regex.Split)
• Add new overloads of MatchesToken() that return ParsedElement instead
  of just matched length
• Add AnyCharTokenPattern — matches any single character
• Add ParserTokenPattern — allows nesting one parser as a token in
  another parser
• Add StringSegment struct — lightweight zero-copy string slice with
  lazy cached substring
• Add SemanticException — structured exception with AST node reference,
  source position formatting, and nested children errors
• Expose PositionalFormatter.Format() as public static method

Bug fixes:

• Fix startIndex in RepeatParserRule, SeparatedRepeatParserRule,
  SequenceParserRule and all combinator token patterns: initialPosition
  now correctly set to first child's startIndex instead of parsing start
  position, which fixes wrong offsets when SkipStrategies are used
• Fix LiteralCharTokenPattern boundary check (position >= barrierPosition)
• Fix RegexTokenPattern null-safety in ToString()
• Fix CaptureTextTokenPattern: set intermediateValue directly on child
  element instead of creating new ParsedElement
• Fix SkipWhitespacesTokenPattern: return child result directly instead
  of wrapping in a new ParsedElement with accumulated whitespace offset

DX improvements:

• Replace hard casts (T)value in ParsedRuleResultBase with soft 'is T'
  checks that throw SemanticException with detailed context instead of
  InvalidCastException
• Same for GetIntermediateValue, ConvertValue, GetParsingParameter

Tests:

• Add ReplaceAllMatchesTests (10 test cases)
• Add ScanTests (14 test cases)
• Add SplitSegmentsTests (8 test cases)
• Update CombinatorTokensTests and KeywordTokenTests to use explicit
  out variable types