Add `known.empty` to `emptyDrops`, to facilitate cell calling on multi-modal data

[amcdavid]

With citeseq data, it might be desirable to estimate the ambient pool of citeseq using the mRNA totals to select putative empty drops, rather than statistics calculated on citeseq totals. This PR adds an argument that allows specifying the assumed-empty droplets by integer index, which are then treated as ambient otherwise. To do this, I refactored .get_lower to return an index of ambient droplets, rather than the threshold on the totals. Thus encompassing by.rank and lower as well.

We're passing R CMD check on bioconductor devel, and I added a couple of unit tests to cover the new argument.

[jonathangriffiths]

Thanks for this PR by the way - apologies for my slowness in getting to it. I will do so soon!