Handling skymaps with NaNs

[physics-png]

Occasionally skymaps will have NaNs or infinites placed in pixels where the minimizer fails to converge:

Currently, FRA does not handle these situations well and instead will place a high probability value in these pixels, rather than 0 or the minimum value on the skymap. FRA should be modified to handle this edge case such that we are either warned there are NaN pixels on the sky, fix the NaN pixels itself and renormalize the map, or perform a combination of the two.

[physics-png]

Note: in addition to counting how many NaN pixels there are, I recommend calculating the total probability of all not-NaN pixels and comparing to the 1.0 expectation.  because not only the number of pixels but also whether they are in high probability or low probability regions matters.  iirc when she did that she found a small (<1%) deficit compared to a total probability of 1.0 (although I also don’t recall or know how the original sky scan calculation handles this, i.e. how does it normalize the pixels in the first place in the presence of the NaN pixels)