fix: unicode token boundaries for non-ASCII “Other” scripts

[Konf]

This PR fixes tokenization for all non-ASCII scripts that go through SegmentScript::Other, e.g Cyrillic languages, Greek, Armenian, etc.

Bug summary

• DefaultTextNormalizer::normalize had explicit split logic for ASCII (normalize_ascii_split), but non-ASCII Other segments were processed as a single token via normalize_span.
• This missed token boundaries in many languages that rely on whitespace/punctuation separation.
• As a result, non-ASCII Other text was normalized as one full segment (for example, a phrase with spaces/punctuation), which made search queries fail.

Fix

• This PR adds normalize_unicode_split function that acts like normalize_ascii_split, but for unicode strings and wires that function for SegmentScript::Other text normalization.
• Now Other text is split into Unicode word-like spans before normalization, so terms are indexed and matched independently.
• Also, 4 tests were added to cover and reproduce such issues and don't allow regressions in future.