Sampling irrecoverably fails for every node when one node fails

[Shaun Read]

https://www.getguesstimate.com/models/16277

When sampling from a node fails (such as a binomial with too large an N), all the other nodes which use a distribution fail as well and continue to fail (throwing up a syntax warning which isn't correct).
Once this happens you can no longer use the model and you have to start again.

I will make a different issue for the binomial sampler which definitely should be able to handle these numbers...