fix: correct Tamil superscript tokenization and subscript indexing (#57)

[vm75]

Description

Resolves #57.

• Fix isolated superscript/subscript token absorption by allowing partial string lookahead consumption in InputReader.
• Implement matchTamilMarker helper to decode canonical phoneme offsets and marker byte lengths for superscripts and subscripts.
• Correct subscript UTF-8 indexing from dividing by 2 bytes to exact 3-byte offsets.
• Add bounded phoneme offset indexing to prevent out-of-bounds mutation on eligible stops and Ja.
• Ensure subsequent preserved characters (whitespace, punctuation, text) are retained intact.
• Separate marker prefix matching from full-span marker verification in isTamilMarkerSpan / isTamilEndOfWord to prevent word-boundary regressions in TamilWriterHandler.
• Add comprehensive component tests in test-handlers.cpp and end-to-end fixtures in test-cases.json.
• Synchronize and regenerate WASM and JS distribution artifacts.