Re-implementing rowsum/colsum for HDF5Matrix

[PeteHaitch]

I just noticed that bsseq has been broken in devel for a while due to the removal of the rowsum,HDF5Matrix-method (which I okayed back in 1a6a44f#comments).

I now remember why I had this method: if the data are so large that they need to be stored in HDF5 then the result of rowsum()/colsum() may itself be so large that it can't be stored in memory.
Therefore, the method I had allowed for this by allowing the result to be written to an HDF5RealizationSink (it also allowed me some extra control over the result, such as writing 2 separate rowsum()s to the same file as in PeteHaitch/DelayedMatrixStats@27712d7).

Would it be possible to have a rowsum()/colsum() that supports realizing the result to a RealizationSink as it goes?
I will try to find time to do this if you're okay with it.

[hpages]

I see. I didn't realize you had a use case for writing the result of rowsum() or colsum() to a file, sorry. I will implement this.

[lshep]

Any chance of getting this change sooner than later so bsseq can pass on the devel builder before the release? Its been broken since 2/01/19

[PeteHaitch]

Hi @hpages, will you have time to do this before the BioC 3.9 deadline or should I backport the functionality to bsseq as an internal helper function?