Solved rsample bug

I increased the atol parameter from 5e-5 to 1e-4, in order to avoid incorrect values (e.g., inf, NaN) in the rsample method.

Solution to most KL divergence bugs

I noticed that kl_truncgauss_truncgauss method would sometimes return incorrect values for the KL divergence. After extensive testing, I found that the formula used to calculate the KL divergence had a typo, where the first two mu2 and mu1 terms needed to be squared.
Now the KL precision is much better, although it still returns incorrect values for some TG parameters.
I tested the precision of the mean, variance and log_Z calculation in these cases, and these quantities do not seem to be the problem. I believe that there are still some issues with the formula used for calculating the KL divergence.

Version 1.3.7

Same release as 1.3.6, but removed debugging code that I had forgotten.

Version 1.3.6

Parallel Truncated Gaussian after solving NaN bug.