Mixpeek integration

Mixpeek is a multimodal data warehouse: files land in buckets as objects, processing pipelines turn them into documents inside collections, and retrievers run multi-stage search (vector, keyword, hybrid, reranked) over those collections. This integration covers that core. Namespaces: create, read, update, list, delete and read usage statistics. Buckets and objects: create and manage buckets, register objects singly or in batch, read, update, list, aggregate and delete them, and drive direct file uploads by requesting an upload URL, confirming it, and doing both in batch. Collections: create, clone, read, update, list, export and delete collections, describe the features they extract, synchronise their schema, apply a taxonomy to the documents already in them, import documents, and trigger processing. Documents: create, read, patch, update, list, aggregate, list distinct values of a field, bulk and batch update, batch delete, delete, validate against the collection schema, update a document access control list, download a document's media, and walk lineage back to the source object or out to everything derived from it. Retrievers: create, clone, patch, list, read and delete retrievers, execute one (auto-optimised, batch, or ad hoc without persisting it), list and read executions with their live status, execution profile and explain plan, run typeahead, and list the retriever stages and feature extractors available. Taxonomies: create, read, patch, list, clone, delete, validate a configuration without saving it, and list or create versions. Tasks: list, read and kill background work. The credential is a Mixpeek API key (mxp_sk_...), sent as a bearer token in the Authorization header, plus the namespace id (ns_...) that scopes the request, sent in the X-Namespace header.