Testing framework

[rjleveque]

Implemented so changes to ClawpackRegressionTest.check_gauges to work better with multiple gauges when OpenMP is used to address #89.

Some changes:

• Assume that all gauges specified in setrun should be checked, so no need to pass in gauge number.
• Don't bother checking sums since all gauge output is saved anyway.
• Read in the gauges, sort by gauge number, and then compare this with the regression_data that has the same form.
• Give a more useful message if the assertion fails.

There's a test case in dev/advection_3d_swirl. nosetests should work with the data in that directory. Try changing a gauge location in setrun.py if you want to see the error output.

I'd suggest we eventually modify the way frame comparisons are done to give a more useful error message and help figure out what might have changed if the assertion fails.

Note: if this gets merged in, Travis tests in amrclaw and geoclaw will fail until the gauge regression data is updated to match what's now expected.

[mandli]

This all looks good to me. I also raised an issue regarding the frame comparisons.