In preparation for the lakeFS v1 launch, change the package to depend
on their newly launched `lakefs-sdk` to keep up to date with their
upstream changes.

This is largely a name replacement only, with the exception of the model
imports now happening from the `lakefs_sdk.models` subpackage.

Mostly a name-only replacement.

Otherwise we end up with a dependency conflict in CI.

This change comes from the move to pydantic, which requires keyword
arguments for its initializations.

This was changed in the lakeFS SDK from the buffered reader used before.

Now, we need to fetch byte ranges, which we were doing in the
`LakeFSFile` already, with every `objects_api.get_object` call.

The main difference, however, is that we cannot just pass increments of
the file system's chunk size, because that will create HTTP416 errors
(request range not satisfiable).

Instead, first fetch the object size from the backend via `fs.info()`,
then guard against the overflow by max'ing it with the next increment
of the chunksize.

We also change a test to purposefully require multiple `get_object`
range requests to see if the logic works as intended.

Command used: nbstripout --extra-keys cell.attachments demos/rain_prediction.ipynb