Further accommodations for changes from DropletUtils::emptyDrops()

- Details of changes: MarioniLab/DropletUtils#118
- For this chunk, the new default of `alpha=Inf` gives poor results. Some quick comparisons given below
  - BioC 3.21
    - 39,028 non-empty droplets with `emptyDrops(counts(hto.sce), by.rank=40000)`
    - 21,374 non-empty droplets with `emptyDrops(counts(hto.sce), by.rank=40000, alpha=NULL)
    - 29,673 non-empty droplets with `emptyDrops(counts(hto.sce), by.rank=30000)`
  - BioC 3.20
    - 21780 on-empty droplets with `emptyDrops(counts(hto.sce), by.rank=40000) (i.e. former default of `alpha = NULL`)
- Added some explanation/justification cribbed from that removed in 76a88c5

Author: PeteHaitch

inst/book/doublet-detection.Rmd

@@ -255,7 +255,12 @@ The barcode-rank plots are quite similar to what one might expect from gene expr
 library(DropletUtils)
 
 set.seed(101)
-hash.calls <- emptyDrops(counts(hto.sce), by.rank=40000)
+# HTO counts tend to exhibit stronger overdispersion (i.e., lower `alpha` in 
+# the `emptyDrops()` calculations), increasing the risk of violating 
+# `emptyDrops()`'s distributional assumptions.
+# Setting `alpha=NULL` estimates alpha using maximum likelihood; see 
+# `?emptyDrops` for details.
+hash.calls <- emptyDrops(counts(hto.sce), by.rank=40000, alpha=NULL)
 is.cell <- which(hash.calls$FDR <= 0.001)
 length(is.cell)