Refactor Lineage class

[michalk8]

TODOs:

• [ ] consider not subclassing numpy.ndarray (related to Lineage class and consolidation of cluster representations with scanpy #759 )
• remove some redundancies (esp. w.r.t. to indexing)
• document/simplify indexing behavior
• fix KeyError when wrong lineage is passed - e.g.
  KeyError: "Invalid lineage name: `Notoch`. Valid names are: `['Early Blastomeres_1', 'Early Blastomeres_2', 'Notochord']`." when passing Notochord2 and Notochord is present

[Marius1311]

How do you want to proceed with this @michalk8 ?

[ivirshup]

@michalk8, as we had discussed before (last year?), I recommend against creating a custom on-disk schema for the Lineage class if you can avoid it.

• It seems like casting things that have been read as arrays to Lineage objects works fine at the moment
• If cellrank hasn't been imported, other libraries will not be able to read in AnnData's containing Lineages
• Libraries from other languages will also not be read these in.

From my perspective, it seems like you'd be giving up a lot of interoperability for something that works as is. In terms of maintenance, on disk formats kinda require you support them forever. APIs can change with a version bump.