Get an upload URL for each file of a session's input.

Name each file and its type, then PUT its bytes to the upload_url that comes back, straight to storage and in parallel if you like, with the same Content-Type as its mime_type. There is no size field and no size limit here: one PUT takes a file up to 5 GB. An upload_url is good for an hour. Pass each id as a {"type": "file", "source": {"type": "file", "file_id": id}} block in a session's or a follow-up's input. When the session starts, each file's bytes are checked against its type and its size is read from storage; a file never uploaded is refused there with 409 ATTACHMENT_NOT_UPLOADED. At most 50 files per call. Rate-limited to 30 calls a minute per caller.