improve argparse usage
    change --host-url to --fatcat-api-url
    add 'help=' to all CLI sub-commands; improves --help output
    do ArgumentDefaultsHelpFormatter everywhere
        parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)

Try stubgen for type annotation::
    stubgen -m fatcat_openapi_client -o stubs/
    stubgen -p fatcat_openapi_client -o stubs/

- schema.org metadata for releases

additional tests
- full object fields actually getting passed e2e (for rich_app)
- modify existing release via edit mechanism (and commit)
- redirect a release to another (merge)
- update (via edit) a redirect release
- api: try to reuse an accepted edit group
- api: try to modify an accepted release
- api: multiple edits, same entity, same editgroup

views
- oldest un-merged edits/edit-groups
- changelog more like a https://semantic-ui.com/views/feed.html ?
- instead of grid, maybe https://semantic-ui.com/elements/rail.html

backlog
- refactor all the "fatcat_*.py" top-level files in module __main__.py files. eg: "python -m fatcat_tools.whatever"
    => or, better, just call individual files directly instead of __main__.py?
- make debugbar really optional (don't import unless we're in debug mode)
- codespell: codespell --quiet-level=2

Kafka
- should refactor all the handlers into one place (DRY) (eg, kafka_fail_fast)
- harvesters could have shared producer setup/creation
- broker timeouts after a minute in all cases should at least log if not crash
  (eg, changelog worker)
- tic/tok increment counter for lower latency but non-verbose logging?