Multi-topic person-grain discovery: given up to 25 topic_ids in one
call, return the people (as HEMs) showing intent across them. The fan-out
runs in parallel server-side, so a 25-topic request is about as fast as a
single topic. Each HEM is a deterministic match key (sha256 of a
normalized email); raw emails are never returned.

group_by selects the response shape:

• entity (default): one row per person, annotated with which of the
  requested topics they show intent on (topics) and how many they
  matched (topic_count). Set min_topics to require a match across at
  least N of the topics (default 1 = union). This view is
  cursor-paginated.
• topic: the raw per-topic ranked lists, one bucket per requested
  topic.

min_score and min_perc_score filter each topic's results.
per_topic_limit (default 1000, max 5000) caps how deep each topic is
scanned. min_topics, page_size, and cursor apply only to
group_by=entity.