rliger Release v2.2.0

• Implemented highly efficient on-disk iNMF that scales to a million cells using
  slightly more time than in-memory version, requiring only laptop-level memory.
• Added 10X H5 data and H5AD loading function that loads the data into regular dgCMatrix
  in memory or the DelayedArray representation backed on disk, the latter is used
  for on-disk iNMF implementation.
• Added selectBatchHVG() which implements another HVG selection strategy, credit to SCIB
• Adding suggestK() back with new methodology
• Clarified optimal runGOEnrich() workflow and added fold enrichment metric in
  the returned result
• Fixed important bug in online iNMF scenario 2
• Fixed multiple problems related to ATAC analysis
  • Fixed Wilcoxon rank-sum test bug when using ATAC peak counts
  • Fixed gene coordinate parsing bug from BED file
  • Optimized peak parsing speed