Cut long text into sentence-aligned chunks under a token limit, with overlap, exported as JSONL
Cuts land on sentence boundaries; a sentence longer than the limit is split by characters. Overlap carries whole sentences from the end of one chunk to the start of the next, so you get a few more chunks. The exported JSONL holds one { id, tokens, text } object per line, ready for an embedding script. Limits are estimated tokens, so real counts drift slightly.