histograms for a group of neighboring runs instead of one run

[c-dilks]

This sounds good. RG-L once discussed creating histograms for a group of neighboring runs instead of one run, similar concept to moving average, when people agree it's better than increasing the mon-scheme sample size per run.

If it is easier to do in qtl-histogram stage, I may have to request for PR to enable an option like last-run. But first let me think about the best way to realize this in the qtl-analysis.

Originally posted by @Sangbaek in #495 (review)

[c-dilks]

My first thought (unlikely the best, but might as well say it): add a script, let's say qtl downsample, that

• takes a number N that is the number of runs to group together
• reads all the HIPO files for a group of N runs, and does the equivalent of ROOT's hadd
  • we might have a hadd-like method in coatjava, but it's likely not hard to write our own, which may be better if we need to mutate plot names or titles
  • assumes all histograms make sense when hadded... some may not...
• produces new histogram files, in a similar file tree structure we already have (maybe even giving it a new dataset name to make a new top-level output dir)
• the "run number" of some group of runs could be the first run, or middle run, or something, since qtl analysis wants 1 run number per input HIPO file
• qtl analysis should "just work" when reading the output files from qtl downsample

Alternatively one could hack qtl histogram, but that script is already pretty complicated...

[c-dilks]

Also, it wouldn't have to be written in bash, if it's a new script; I kind of regret choosing to write most of the bin/qtl* scripts in bash...