First steps for the implementation of the new dataset interface: 
- Removed Pipeline class
- Added Workspace singleton to hold pipeline name, base_path, etc. ..
(shouldn't be the focus of this PR)
- Moved `Pipeline.read(..)` to Dataset class

Implements the new workflow including initialising datasets from
manifest.

- removes workspace again, using working_directory argument now to
enable something like `fondant run ... --working_directory gs://...`
- removes base_path from the manifest and changes the field location to
absolute paths
- changes the executor to generate absolute field locations
- add `Dataset.read()` for initialising dataset from manifest path
- adjust compiler code


Try to create sketch out the flow and changes. Still have to fix the
test cases and test the changes end to end for all runners.


![image](https://github.com/ml6team/fondant/assets/15777729/8c330daf-2076-493c-aea2-b7d3f85884b8)

Co-authored-by: Matthias Richter <[email protected]>