Update DOFArray flatten and unflatten

[alexfikl]

Following up on inducer/arraycontext#91 (comment).

At the moment this just updates the docs in the dof_array flattening functions to also mention the ones in arraycontext. There's a few of these: flatten, unflatten, unflatten_like, flatten_to_numpy and unflatten_from_numpy; and all of them just work on DOFArrays and leave the rest of the container alone.

How many of these should be deprecated? At least the numpy ones have some nice uses

• in some tests, but these can probably be replaced with some reshaping, e.g.

nodes = thaw(discr.nodes(), actx)
nodes = np.stack(flatten_to_numpy(nodes, actx))

• pytential uses unflatten_from_numpy in a couple of places. Not sure any of those actually flatten anything that's not a DOFArray though.

• pytential also uses flatten quite a lot to talk to sumpy. This could be replaced by arraycontext.flatten + reshape too, to make it (3, nparticles), I think, since sumpy supports both types of inputs.

[inducer]

I don't see a principled reason to keep any of them. They seem kind of redundant now that there's a documented way to flatten given just an array context.

[alexfikl]

I don't see a principled reason to keep any of them. They seem kind of redundant now that there's a documented way to flatten given just an array context.

At least for the tests in meshmode it was impressively simple to replace them with the arraycontext versions 🎉. Should we also remove them from the docs?

[inducer]

Yep, I say let's remove them from the docs as well.

[alexfikl]

Yep, I say let's remove them from the docs as well.

Removed in 783686d.

[inducer]

Thanks! Spotted one wrinkle, otherwise LGTM!

[inducer]

Thx!