HVF does not account for sample differences

The ATAC analysis pipeline currently places the HVF (highly variable feature; analogous to RNA HVG) computation at the end of the QC notebook. The idea was to use HVFs for filtering. However, the data isn't normalized at this point, so HVFs could arise from differences between samples, such as sequencing depth. Move the HVF step between normalization and dimension reduction in the 3rd notebook (same as RNA).