Use arraycontext in tests

[alexfikl]

Ports all the tests to the arraycontext infrastructure.

This is not on top of #118 because it can go in without the rest of sumpy being ported.

[alexfikl]

After writing most of this, I realized that there's some small conflicts with #118. I'm happy to resolve those if this gets in.

[alexfikl]

@inducer This should be ready for a look!

Most of it is just tedious, but there are a few things worth mentioning:

• This inherits the array context from boxtree.PyOpenCLArrayContext (instead of the one from arraycontext).
• There were a couple of tests that used np.random and failed when changing the seed in test_kernels.py.

[inducer]

• There were a couple of tests that used np.random and failed when changing the seed in test_kernels.py.

Do you remember which ones those were? Were they just random tolerance botches or something you thought was indicative of a deeper issue? (If the latter, could you file an issue?)

[inducer]

This looks good to me, and it's an all-around improvement. Thanks!

[alexfikl]

Do you remember which ones those were? Were they just random tolerance botches or something you thought was indicative of a deeper issue? (If the latter, could you file an issue?)

@inducer Looking through the failed CIs (e.g. https://github.com/inducer/sumpy/runs/7933601203?check_suite_focus=true)

• test_kernels::test_translations: failed with assert ([-0.19, -0.8] < -0.2).all()
• test_kernels::test_p2e2p: failed with assert 3.3 > 3.5

so I would say it's just a case of fixing the tolerances to the previous seeds, but I don't know enough about what the tests are actually testing to be certain.