Implement a second iteration to the pipeline

[nictru]

After creating a basic atlas and manual annotation, we can subset to certain subpopulations (e.g. Myeloid cells) and run the Neighbors-Leiden-UMAP section again on only this subset.

The resulting UMAPs can then be explored using the embedding switcher, since switching between integration methods does not make sense for this.

A reason for this is that if there is a very high number of cells, closely related cell types are not clearly distinguishable in the UMAP any more, although in the clustering it kind of works