Renormalize KDE PSF

[jvansanten]

For the likelihood to be unbiased, the signal and background spatial PDFs need
to be normalized on the sphere. Setting S/B to zero outside the spatial box
explicitly breaks this normalization if any of the probability mass is outside
this area. The KDE PSF, this can be on the order of 2-5% depending on energy,
which appears to be enough to produce ~50% biases in n_s for a few hundred
events injected over 1000 sources. The same problem exists for the circular
Gaussian (Rayleigh) PSF, though on a much smaller scale.

This appears also be responsible for commonly-observed biases in gamma, since more of the PSF mass is outside the selected area for low-energy events.

Work around this by explicitly normalizing the signal PDF to a circle whose
radius is given by the configured spatial_box_width. In admittedly limited testing, this appears to significantly reduce bias in n_s as well as gamma for high-background stacking analyses using StdMatrixKDEEnabledLLH.

This PR depends on some commits in #440