Adopt uax29 segmenter

Replacing blevesearch/segment. ~2x throughput improvement. Refactor allocations, now ~O(1).

Add tests & multilingual sample text to ensure identical behavior. Known differences from previous segmenter:
- The original segmenter splits runs of spaces into separate tokens; uax29 concatenates runs into a single token.
- The original segmenter doesn’t handle emoji skin tone modifiers, the new one does, attributable to Unicode version update.

uax 29

Author: clipperhouse