Function for integrating datasets using scANVI

We would like to use [`scANVI`](https://docs.scvi-tools.org/en/stable/user_guide/models/scanvi.html) for data integration benchmarking, which requires a list of [AnnData](https://anndata.readthedocs.io/en/latest/) objects and cell type annotations. scANVI does everything that scVI does, specifically integration of datasets, while incorporating cell type annotations that do not have to be complete. Datasets can have annotations or if only one dataset has cell type annotations, that dataset is treated as the reference set and cell types are inferred based on integration.

One thing to note is that in order to use these tools, we might need to obtain a GPU (although it says `effectively requires a GPU for fast inference` so maybe we don't need it for initial setup). As with #16, we would need to convert SCE objects to AnnData, and then follow the instructions for running `scANVI`. The documentation is not the easies to follow, but I believe after model generation, we would get a new integrated annotation and then we would add the updated annotation to the original AnnData. This [documentation has an example of using scANVI for atlas level integration and label transfer](https://docs.scvi-tools.org/en/0.10.0/user_guide/notebooks/harmonization.html).

Read the linked scANVI atlas-level integration and label-transfer example, then compare it with the SCE-to-AnnData conversion discussed in #16. Define the function interface for a list of AnnData objects and optional cell-type annotations, and confirm that completion includes an integrated annotation added back to the original AnnData data.

Written by the indexing model from the issue text.