This is a start for consolidating common functionality scattered
around various extensions into a single implementation (pattern).

This changeset import `iterdir()` from `datalad-gooey`. In contrast to
the original implementation, this new one is using a stricter approach
to types, and overfits less to a dataset-aware use case.

However, it is not meant to be the exclusive implementation, but merely
a start and a place to migrate directory iterators into.

Ping datalad#323

This is trying to be structurally similar to the `directory` collection
implementation.

Only one path type enum, only one item dataclass (for now).

Now also for link targets.

Using a pre-crafted (300 byte) tarball that is put on github, and is
downloaded once per session.

Also include a check for these requirements that is executed in
the error case (no performance critical), to inform developers
about obvious implementation issues.

Closes datalad#348

This is (also) an alternative approach to `add-archive-content`.

In comparison to the former, this is largely metadata driven, and
works without (local) extraction of a tarball. This saves storage
overhead, and makes it possible to run some parts of the ingestion
pipeline on a remote system.

Closes datalad#183