The main reason for this update is that we had duplicate copies of
histograms in our output file, which doubled the size of histogram
files and meant that user doing a simple loop over all files in the
output file would process histograms file.

In order to make the whole mechanism more robust and testable I moved
all the output stream handling into the class `OutputStreamData` and
reworked that so that it looks and behaves like a real class, with
information hiding, etc.  And I added unit tests that made sure that
output objects indeed make it into the output file and only get added
once to that output file, plus make sure that all the special logic
and handling indeed works properly.