User-data validation and CAS output summarization for scanpy visualization

[mbabadi]

The following branch contains helper methods for user-data validation, CAS output summarization, and visualization of CAS results using scanpy:
https://github.com/broadinstitute/cell-annotation-service-client/tree/mb_cas_cli_scanpy_demo

Tasks:

• Incorporate the prototyped user-data validation method with the annotate function. The prototyped implementation is here:
  https://github.com/broadinstitute/cell-annotation-service-client/blob/8f4433eca90479225d496b41af7f1259893bacb4/src/cas_client_helper.py#L12
  This implementation is not memory efficient and is not ideal for large datasets since it makes a new copy of the raw data matrix w/ reordered feature columns:
  https://github.com/broadinstitute/cell-annotation-service-client/blob/8f4433eca90479225d496b41af7f1259893bacb4/src/cas_client_helper.py#L61
  A better implementation must apply matrix reformatting on chunks. The necessary indices (gene_id_intersection_casp_indices and gene_id_intersection_adata_indices) can be recomputed once.

• Get rid of:
  https://github.com/broadinstitute/cell-annotation-service-client/blob/8f4433eca90479225d496b41af7f1259893bacb4/src/cas_client_helper.py#L65
  which is a temporary workaround for PCA inference expects the input anndata to have integer indicies cellarium-cloud#35.

• Make a separate utility submodule for postprocessing CAS output (the last three methods in cas_client_helper.py. If we change the current CAS output, these methods will break; and we are very likely going to need to change the CAS output many times. It would be useful to version CAS output and assert compatibility (will make a separate issue for this).

• Polish the notebooks (just the two PBMC notebooks)