Produce bands containing experimental shape uncertainties

[kreczko]

(following TOP-12-028).

OK, this is a hard one (relatively speaking).
Problem 1: not all of of our systematics are evaluated pre-unfolding: matching threshold, scale uncertainty, PDF weights, hadronisation.

In order for this to work we need to ntuple-lise the corresponding samples and run them through AnalysisSoftware.

Problem 2: Hadronisation uncertainty is evaluated w.r.t. two POWHEG samples, none of which is the one used for the central measurement. The way to 'normalise' all uncertainties we would need to create two histograms: errors upper bound and errors lower bound. This procedure will have to be performed for every distribution we want to show.

Problem 3: If we want to include the PDF uncertainties, this will take a lot of processing time for not much value (??). I suggest we ignore this one for now (maybe even all generator level TTJet systematics).

Proposed solution: create a script that takes our input samples and our systematic samples after AnalysisSoftware running. It then needs to

1. normalise the systematic samples to luminosity
2. evaluate hadronisation uncertainty and scale to central sample
3. evaluate the PDF uncertainty and summarise the result to a set of PDF up and PDF down
4. summarise all of the above to a single ε₊ and ε_- value

The above output can then be used to construct the error bands.