Automatic handling of Z dimension

[angus-g]

The initial design of the library was done with 2D datasets. While we don't necessarily want to perform 3D advection, there should be better handling of datasets with a non-trivial Z dimension. I can imagine that for a small enough dataset, we'd be able to seed particles on each distinct Z slice, and advect all particles in 2D, simultaneously. However, this may be infeasible in terms of memory for a large enough dataset. In this case, would it make sense to loop over all slices (this should probably be an outer loop, so that we can take advantage of data for the current slice being cached as we advect back and forth.