Longest-match word segmentation

Greedy longest dictionary match aligned to syllable boundaries.

• Matches cannot split a syllable.
• No merge across whitespace (token adjacency required).
• Grammatical suffixes (သည်, များ, …) stay separate tokens (myPOS-style).
• Digit + classifier may fuse (e.g. သုံးခု).

Deterministic and fast; not globally optimal. Larger lexicons or statistical engines can plug in later via tokenize.engine.