resurrect option to use fitting in HillasReconstructor

[kosack]

Right now, HillasReconstructor loops over pairs of lines, finds their intersection point, and then does a weighted average of the sum (actually doing the cross-product, so the weight is by sin of the angle). Here the weights are for pairs of images.

In MARS (and very old versions of HillasReconstructor), there is the option to use a least-squares fit, which allows one to weight by image, rather than pairs, which can be useful if we want to use weights based on e.g. the MISS parameter.

Re-implementing this fit would be useful, at least as an option, so that we can try different image weighting algorithms.

e.g. something like https://stackoverflow.com/questions/52088966/nearest-intersection-point-to-many-lines-in-python/52089867