batch evaluation

[JunnanZ]

Just a small correction to the readme file: the interpolation function I wrote already handles evaluations at multiple points, as indicated by the docstring here, without using vmap. It's all vectorized because map_coordinates is vectorized.

A quick test shows that it is faster than eval_linear when the number of points to evaluate is large (>100000) but slower when the batch is small. I think this is because the GPU with a large number of cores is better suited to handle large parallel computation.

Please see the attached test file, which is adapted from Smit's notebook.
linear_interpolate.pdf

@jstac @shlff @Smit-create

[JunnanZ]

Also, the dimension 4 is hard coded for Jax to compile. It can be changed to any number without changing anything else, as I did in the notebook attached above. So, a quick way to generate jit compiled interpolation functions that work on other dimensions is to write a function factory that wraps my functions with n as an argument.