* previously it seems like it was missing after pipx install

* also, remove `MANIFEST.in` because that's not used in the poetry build
  backend: https://python-poetry.org/docs/basic-usage/

* wrap the file IO in a context manager to close it once it's not needed
  anymore
* call `set_content()` only on the file in question, not all files for
  the draft (that have likely already been committed)
* override the `_data` property of the outgoing stream to prevent
  the requests library from doing its magic

* the file paths are now based on the container id and file name
* if no base directory is specified, it is based on a random temporary
  directory
* the path for the base directory is based on `tempfile.tempdir`, which
  can be influenced e.g. through the $TMPDIR environment variable

* for now, only `download()` is supported

* also, update the top-level description

* if the `interactive` flag is set, actually query the user
* if it's just a single file anyway, just download that

* missing keys in the dictionary comprehension turned it into a set...

* as it turned into a broken link on PyPI

* the previous logic was faulty because it added "/api" as suffix
  instead of prefix
* even using it as prefix caused issues, because the client expects a
  pure API endpoint without any trailing path

* these functions only act as wrappers for actual (one-shot) services
  for simple use-cases and don't need state handling
* remove the `archive_data()` function for now, as that's currently not
  a relevant use-case

* knowing the container ID helps with subsequent operations if the
  container didn't exist before
* also, allow the recid in InvenioRDM to be omitted (in which case a new
  draft will be created)

* move the common ones to a new utils module
* also, allow the InvenioRDM URL to omit the protocol