(bug): `pbmc3k` tutorial is not reproducible

[ilan-gold]

To Reproduce

Run https://scanpy-tutorials.readthedocs.io/en/latest/pbmc3k.html with the current scanpy master branch and the results of marker gene plotting do not match the table. For example IL7R is not ranked highly in the first group (notwithstanding the fact that it is filtered out by the highly variable genes selection). It seems that even in the rendered version, this is not happening.

I also get a different number of clusters as a default.

From what I heard, this tutorial will be replaced anyway but figured I'd register an issue.

[TT-Omics]

I am also getting different markers - the plots are significantly different. My results really start to diverge once start the paga, umap, and leiden clustering. They do Leiden last, but the code will not run unless you do Leiden first... am i missing something?