Refactor, DensityEstimatingSource

[JelleAalbers]

This is a refactor to reduce complexity and enhance extensibility. Well, that says pretty much nothing :-)

• A new Source class DensityEstimatingSource can be used when you want to estimate the PDF from events, but don't want to generate events with the simulate() method. For example, load events from some data file, or run an external Monte Carlo.
• PDF interpolation / morphing in likelihood space is split off from LogLikelihood into a separate file pdf_morphers.
  • Different morphers (currently GridInterpolator and RadialInterpolator) have their own classes.
  • You can control which morpher is used by the morpher option to thelikelihood_config option of LogLikelihood.
  • Similarly, use morpher_config to set morpher-specific config options.
  • Source and Model are now properly separated, model no longer passes its own self to source.
• The individual-source parallelization has been removed. I don't suppose we'll need it again: you usually want to generate several models in parallel (for shape uncertainties).
• Various other private / internal stuff has been moved around.

[coveralls]

Coverage increased (+8.0%) to 83.562% when pulling 4de1159 on refactor_ into 1ef5f98 on master.

[coveralls]

Coverage increased (+7.9%) to 83.436% when pulling 98dbc17 on refactor_ into 1ef5f98 on master.