ListDocuments lists documents within one bucket, with optional search, filtering, and sorting.

The pool holds unrelated corpora side by side, so bucket is required and a listing
never spans more than one of them.

Setting search turns this into a retrieval call rather than a listing: it runs a
hybrid semantic and lexical search over document content and returns the passage
that matched in each item's content field. Ranking is by relevance, so offset, sort
and order are ignored, and total counts what is returned rather than what exists.