Sparse jac

[spcornelius]

Addresses #89.

From a design standpoint, I was conflicted about whether to:

1. implement the sparse jacobian as a separate callback/set of expressions (cf. jtimes, dfdx, etc.)
2. Overload the behavior of the jacobian callback/symbolic expressions

I opted for the latter, mirroring how pycvodes currently accommodates dense vs. banded vs. sparse jacobians. The drawback is that a given SymbolicSys/ODESys is implicitly tied to certain integrators/linear solvers at time of instantiation, rather than existing as an independent construct usable with an arbitrary integrator...

[bjodah]

Great. I see the dilemma, it's not too bad since we have SymbolicSys.from_other which allows converting back and forth by setting sparse=True/False. When I tried this branch out I did it in a notebook, and tested this. I took the liberty to check it in as well.

[bjodah]

This is now in pyodesys-0.12.3