UMAP coordinates between core and extended atlas

### Description of the bug Hello @nictru , I want to report what I think is a bug in the extension workflow of scDownstream. I have a set of datasets from different sequencing technologies that I'm using to build an atlas. The core atlas looks perfect, but not the extended one. While the pipeline runs trough, the datasets from the extended atlas seem to cluster all together, separately from the rest. While the cell type differences are visible, it's almost as if they cluster in their own mini-umap, with a different coordinate system from the main one. I can't provide you with public screenshots here but I can share them privately with you. Keep up with the amazing work, and thanks for your help! ### Command used and terminal output ```console ``` ### Relevant files _No response_ ### System information _No response_