Add tests of distance marginalized likelihood

I wanted to see how much adding in tests of the distance marginalization will slow things down.

Previously we had not used this because it was too expensive to do. I think it should now be better because:

• we now cache the lookup table so it only has to be built once with and without phase marginalization.
• building the lookup table was sped up recently by vectorizing an inner loop.

I also noticed during the testing that we were tripping one of our own warnings about using a deprecated method, so I fixed that.