Investigate whether current KNN imputation default (FKNNI) works for all sparseness and AnnData objects

[Zethson]

Description of feature

We ran into cases in the past where users suddenly reported that it didn't impute everything and we need to understand this.

• Generate a loooot of different AnnData objects with different numbers of missing data etc. A LLM can easily do that for us.
• Run KNN imputation with backend=fknni (default) on all of them -> check whether any of them break

[eroell]

We ran into cases in the past where users suddenly reported that it didn't impute everything

Does this relate to the same reportings as #734?

[Zethson]

Yes! Sorry this is a duplicate now :(