[Feat] Add support for index tricks in autograd numpy

[fjosw]

As requested in #623 I had a look at adding support for np.mgrid and a few other index tricks. Just importing them in numpy.__init__ makes them available but I'm not sure how much value that actually adds. What is your opinion @agriyakhetarpal ?

[j-towns]

The functions

diag_indices
diag_indices_from
fill_diagonal
ndenumerate
ndindex
ravel_multi_index
unravel_index

are all already there, so we don't need to add them. They are wrapped because they are callable, which happens here.

The only ones we potentially need to add are

index_exp
mgrid
ogrid
s_

They don't get wrapped because they accept their arguments as indices. It should be possible to wrap them in a similar way to r_ and c_, which are done here. If you want to try implementing that I'm happy to review a pr.

Probably best to do it in autograd/numpy/numpy_wrapper.py, not in autograd/numpy/__init__.py.

[agriyakhetarpal]

I'm sorry; I just realised I had noticed this and not responded earlier. I do not have much to say here – I would echo Jamie's suggested changes. It would be nice to have a few test cases as well.