PDFs with all zeros in BPZ

[sschmidt23]

I was playing around with the evaluation demo notebook and using BPZ in place of FlexZBoost, it threw an error because a handful of objects have weird colors and small errors, so the chi^2 fit is terrible, and the likelihood=e^-chi^2/2 ends up at 0.0 for all redshifts. This is obviously a problem when you try to normalize the PDF in qp, and the notebook threw an error. So, I should probably add a check to BPZ_lite to set these to a non-zero uniform value rather than 0.0.

[eacharles]

This seems like it should be moved to an issue in qp for dealing with flags in general, which probably just means agreeing on a convention for how to use qp ancillary data to flag bad distributions.

[eacharles]

Second comment, this also related to the flag conventions.