Address Instability and Convergence Issues with the GeoClaw Riemann Solver

[mandli]

It was recently noticed that an instability now occurs due to changes in the GeoClaw Riemann solver. Addressing this in the short term, we are rolling back the suspected changes that caused the problem (#187).

The larger issue is that the GeoClaw Riemann solver needs to be better understood so that changes that need to be made can be applied with better knowledge of the consequences. This may include a refactor, better tests, and a simplification of what the solver actively does. Possible actions:

1. Refactor and simplify the current solver.
2. Create a test harness so that individual Riemann problems can be run.
3. Create a Jupyter notebook that demonstrates what the solver does using the capability above.
4. Create a suite of representative tests that can be tested directly with the Riemann solver that does not require the full framework of GeoClaw to run.

These are just suggested courses of action for right now and any other thoughts should be included here.