Possible to mask channels in NLL evaluation

[ajgilbert]

• CachingSimNLL can now be given a list of RooAbsReals (one for each
  component pdf) which cause the corresponding channel to be skipped in
  the NLL evaluation when the value is != 0.
• RooSimultaneousOpt can also hold on to this RooArgList, and will hand
  it over to the CachingSimNLL in the createNLL step
• These parameters can be defined by the user in the PhysicsModel via
  the new method getChannelMask(bin). By default a RooRealVar named
  mask_[CHANNEL] will be created with a value of zero and set
  constant.
• The RooArgList is created and passed to RooSimultaneousOpt after the
  latter is created in ShapeTools.py
• It is then possible to mask a channel at runtime with
  --setPhysicsModelParameters mask_CHANNEL1=1,mask_CHANNEL2=1,...

Only done basic testing of this so far with single fits in MultDimFit, but seems to work as intended: masking all but one channel gives identical results to just fitting a datacard with only that channel. But should check a wider range of cases to make sure there's no odd behaviour.

[adavidzh]

This looks useful. I think it is just missing some verbosity. I would say that in case a mask is set, then it could dump all the masked/unmasked channels at -v 3 or so.

It also seems that it will silently fail in case a non-existent channel name is passed. Is that right?

[ajgilbert]

Getting back to this... decided to make it off by default for now, can be enabled with text2workspace.py --channel-masks.... A list of all the mask vars and their values is now printed with -v >= 2. It will indeed fail with a non-existent channel name, however since it's all going via --setPhysicsModelParameters the usual Warning: Did not find a parameter with name blah does get printed.