Fix handling of prior terms in ExactMarginalLogLikelihood

[saitcakmak]

This fixes an issue originally discovered in pytorch/botorch#1259. The bug relates to handling of prior terms in https://github.com/cornellius-gp/gpytorch/blob/master/gpytorch/mlls/exact_marginal_log_likelihood.py#L42-L43

When training batched model of n_batch batches, the res term (which is the log probability of observations under the prior) has shape n_batch. The prior term prior.log_prob(closure(module)) also has a shape that is n_batch x (potential other dimensions). When we add the sum of all prior terms via res.add_ we end up adding each prior term to each one of the terms in res, when they should only be added to the term corresponding to the same batch. This is corrected by replacing the sum() with a series of sums that reduce the prior term to have the same ndim as res.

Test plan:
Units

cc @Balandat

[saitcakmak]

This seems to be the same issue as #1318, so that should also get fixed.

[Balandat]

Thanks!