with FMMLib-based expansion order finder

Closes #39

- [ ] Point requirements.txt back to pyfmmlib master after pyfmmlib!15 is merged

Tests the most important functionality (P2E, E2P, E2P).

Adapted from the 3D paper experiments.

* renamed BlockIndex to BlockIndexRanges.
* renamed `view` to `take`, since the methods allocate and would cause
confusion.
* reduce the number of queues being passed around.

* created a new class that holds 1D block index information.
* add a few convenience functions to hide the implementation details.
* make everything work with cl arrays and add some functions to do a
mass get() on all the arrays.

* replaces mat / scalar with mat * scalar**-1.
* replace list argument to subs with a dict

(Although there are pending changes to SymEngine for the above two
issues, they aren't released in a version yet.)

* Removed deep as a kwarg to expand(). It doesn't do what we want
anyway.

* Disabled caching for tests in test_kernels. This exercises SymEngine
code generation every time the tests run. The execution time for these
tests without caching is about 6 minutes for me, so the overhead is
acceptable.

At least on my branch of SymEngine (v0.3.0), this test fails because
it greedily evaluates the Subs nodes.