Implement likelihood scans

[davidwalter2]

Can we make use of constrained minimizers (as done in combinetf1? https://github.com/bendavid/HiggsAnalysis-CombinedLimit/blob/tensorflowfit/scripts/combinetf.py). Maybe gain from other improvements in tf2 and/or scipy.

[davidwalter2]

1D likelihood scans and likelihood contour scans are now implemented and work in principle. There are a couple of things to work on:

• In several cases we use SR1 for the moment but hessian or hessian vector products could be better.
• For the scans it's not clear if constraints are needed or if the parameters can be freezed to some value instead.

There are implementations for 2D likelihood scans, the 2D contour scan isn't working at the moment. The 2D likelihood grid scan is in principle but could probably be optimized.