get object

Downloads an object to a file. The object is streamed to disk, so a large file costs no extra memory.

Downloads `bucket`/`key` and emits it on the `file` pin as a normal blueprint file descriptor : {filename, path, type} inside an ARRAY : written into the entity's temp tree. Streams straight to disk through the atomic writer, so the file never exists under its final name while incomplete and nothing is buffered in memory. The stored filename is a random 12-character prefix plus the key's basename, so two flows downloading the same key cannot collide. Refuses an object over 5 GiB (OBJECT_TOO_LARGE) rather than truncating it. `last_modified` is an RFC-3339 string, or null when the server sent none. A MISSING object is a FAILURE here (success=false, error NoSuchKey) : use s3connection_head_object when absence is an expected branch rather than an error. Other failures: NO_CONNECTION, MISSING_BUCKET, MISSING_KEY, DOWNLOAD_FAILED, COULD_NOT_WRITE_FILE, or an S3 code such as AccessDenied.