Fix VNNGP with batches (#2375)

* fix vnngp batch mode

* fix format issue

* fix nearest_neighbor_variational_strategy batch compatibility

* fix nearest_neighbor_variational_strategy type error

* Improve type hints

* Allow x to be broadcasted to inducing points batch shape

* Fix behaviour for training_batch_size == M

Create only a single taining minibatch with all the data, and use the appropriate kl_helper

* Remove unnecessary settings

* Remove M=k case

* Handle mixed model/inducing point batch sizes with KL helper

* Test cases catch batch VNNGP errors

---------

Co-authored-by: Geoff Pleiss <824157+gpleiss@users.noreply.github.com>