refactor(exporter): reorganize record export flow

[slint]

AI disclosure: I directed this refactor with Codex and reviewed the resulting diff in detail. AI wrote pretty much most of the move, new implementations, and tests. I had to go through many iterations though, referencing the legacy Zenodo exporter module, and getting the right naming and "split" between the different concepts.

• Split the logic from the Celery task into separate files. Some part of
  the design was borrowed from the legacy Zenodo exporter module.
• Produce multiple export formats from one record scan.
• Use a temporary "staging" file(s) to write the archive(s) before creating
  the final published ObjectVersion(s).