File Upload APIs

Routing a large file upload through the same API server that handles ordinary JSON requests works for small files, but it wastes server resources holding open connections and buffering bytes it doesn't actually need to process - the presigned-URL pattern lets the client upload directly to storage, with the API server only ever handling the metadata.

The `upload_url` is a time-limited, cryptographically signed URL granting temporary write access to one specific storage location - the signature encodes exactly what's permitted (this bucket path, this content type, until this expiry) without requiring the client to have any broader storage credentials. The API server's only job is issuing that signed URL and recording the upload's metadata; the actual bytes never pass through it.

A single PUT works fine up to some reasonable size, but for very large files (video, large datasets), a single failed connection partway through means restarting the entire upload from byte zero - multipart upload splits the file into chunks, each uploaded (and retried) independently, with a final step that tells the storage layer to assemble the chunks into the complete file. This makes a transient failure on one chunk cheap to retry, instead of expensive to restart from scratch.