Record free text and let brain extract the facts

The simplest write, and the one to reach for when the caller has prose rather than a structured claim: a chat turn, a meeting note, an email body. The text is captured as an episode, an LLM extracts entities, facts and edges from it, and each extracted fact goes through the same bitemporal resolver as POST /v1/ingest/fact — so extraction is noisy by design and conflict resolution cleans up at read time. Only text is required: contextRef defaults to the chat vertical and emittedAt to the moment the request arrives. userId stamps the per-user memory scope on the episode and every extracted fact and edge. skipped: true means nothing new was recorded — reason says why: empty text, no_entities (the extractor found nothing worth recording), or duplicate (this user already sent this exact text; the document store keys on the content hash and a replay commits nothing new) — a normal outcome, not an error. Prefer POST /v1/ingest/document for anything long enough to need chunking. Source: src/ingest/ingest.controller.ts.

Required scope: brain:write.