Fix bugs in inference

[georgematheos]

@nishadgothoskar this seems to remove the NaN bugs. Inference seems like it works alright, but it does get off eventually -- and I think it isn't honing in closely enough to the right pose. So I think it is a good next step to do as you suggested and add several coarse-to-fine steps.

[georgematheos]

The inference bug arose because the function to propose vertex colors given the other attributes assumed that the latent point projected to a pixel in the image -- but sometimes it would be called on a latent point that did not hit any pixel. The fix here adds code to branch on whether the observed_rgbd is valid or invalid (ie. the token for when the latent point doesn't hit a pixel).