Add a more interesting analysis skim example
Based on https://gitlab.cern.ch/cp3-cms/bamboo/-/merge_requests/203 (hence the WIP status), which is was waiting for !195 (merged) .
This adds a more interesting example of a skim for analysis plots (trigger and two fakeable leptons, as for the ttW measurement), with the leptonMVA precalculated. The same additional branches that are stored in the skim are also added as attributes in the scenario where the input trees are used directly, such that switching between the two is transparent (and a mix of skimmed and unskimmed samples would work).
The other changes are to support a mix of "array" branches and "RVec" branches, which is what is produced by Snapshot.
Few motivation numbers from running this with the ttW signal sample (they mostly show that the MVA dominates CPU usage if you make just a few plots - the takeaway is: it depends on the case, but the difference may be large, so it's probably worth checking):
- plotting without skim: 7kHz
- skimming (with lepton MVA evaluation) 1.5-1.6kHz
- plotting from the skim (ok, few plots): >22.5kHz (not enough data in a file to make the JITting cost at startup negligible; these are the 20% of events that pass the filter, which will be far less for data and backgrounds)