Forward-facing step example

[ketch]

This builds on and replaces #445. Mainly, I've edited the comments to better document this example. There are some unfortunate things:

• Using the Roe solver (and Classic) we get a carbuncle instability after a short time. I understand the solution is to use an HLL solver.
• The approach we're using for internal boundaries seems to work well for Classic, but not for SharpClaw. I think this is because the values used for reconstruction don't make sense after a while. We could implement a different approach, in which cell values inside the step are overwritten in a before_step routine, but there is then ambiguity in what value to write to cells near the corner of the step.

[mandli]

This looks like it's failing with the same problem as before when we were having trouble with our <#> version.

[ketch]

Okay, I rebased on master. Let's see if that fixes it.

[coveralls]

Coverage decreased (-0.47%) when pulling 5d9da81 on ketch:forward_step into 530d7c0 on clawpack:master.

[mandli]

Passing now, merging now.