[QUESTION] How can non-overlapping embeddings be handled smoothly?

[nictru]

In my use case, there is one embedding for all cells and multiple embeddings for subsets of cells, which are non overlapping and of various sizes. Let's assume a structure like this:
Embedding A: 100 cells
Embedding B: 50 cells
Embedding C: 10 cells

After using the embedding switcher to cycle through the embeddings and going back to A, I now have A: 10 out of 50 cells, with no straightforward option to undo this. A hacky solution is to click the subset and undo subset buttons, which basically removes the history of visited embeddings and thus makes all cells in A active again. I don't think this is a great solution.

I have been digging through the source code a bit today, and found out that basically for all embeddings (dimensions) that have been visited, the intersection is built here. I was able to prevent the creation of a history of visited embeddings by commenting out this line.

So now my question is:
Why does it make sense to build an intersection like this? Can this be solved in any other way?

I understand that this is a problem which would not come up if all embeddings have the same size, which is probably the case for most users.