getPosteriorLogDensity method of conjugate samplers doesn't modify target value

[danielturek]

DO NOT MERGE.

Creating PR to trigger testing.

[danielturek]

This PR has passed testing, suggesting the conjugate samplers still function correctly with these changes.

Waiting for @paciorek to determine if this fixes the problem he observed (#924) in the crossLevel sampler.

[paciorek]

Looks good, except I'll note that even in identity link cases we write back into the model even though model[[target]] has not changed. Is it worth checking and only doing this if necessary? Or best just to always do it?

[paciorek]

@danielturek I'd like to merge this in soon. Can you look at the identity link case and see what you think vis-a-vis my comment above?

[danielturek]

@paciorek Apologies that this slipped by me back in August.

I've just updated the crossLevelFix branch to reflect your suggestion: the conjugate sampler will not "restore" the value of the target node, nor call model$calculate(...), when the link is "identity", in which cases they should not have been modified.

Please take a look any time, and let me know if you agree.

Assuming so, I'm fine for you to merge this in.