[BUG] normalize_per_cell can lead to not exact reproducible results with mixed data

[dweemx]

Describe the bug
Applying the scanpy normalize_per_cell function 2 times on the same expression matrix (containing 10x data and RPK SMART-seq2 data) lead to 2 different but very similar results. This bug has also been described more thoroughly here: scverse/scanpy#1612.

To Reproduce
Steps to reproduce the behavior:

1. Configure with these options:

nextflow config -profile tenx,harmony,singularity > nextflow.config

2. Run using this entry point:

nextflow -C nextflow.config run vibsinglecellnf/vsn-pipelines -entry harmony

3. See error:
   No log error. The problem is that the X slot in the AnnData is not reproducible after the normalization step normalize_per_cell between 2 runs. This leads to minor differences into downstream steps (embeddings, ...)
   

Expected behavior
Both should equal. Interestingly, this seems to happen for smaller datasets like PBMC 1k, 3k datasets from 10x.

Screenshots
If applicable, add screenshots to help explain your problem.

Please complete the following information:

• OS: CentOS
• Nextflow Version: 20.11.0-edge.5448
• vsn-pipelines Version: 0.25.0

Additional context
Add any other context about the problem here.