Separate benchmarks from unit tests (add benchmarking suite)

[rkingsbury]

Many of the original unit tests for Solution, especially for get_activity_coefficient, density, conductivity, etc. were written as comparisons to literature data or output from the PHREEQC pitzer.dat model. Moreover, many of the tests are redundant (as unit tests) in that they test the same property against literature at multiple concentrations.

Work should be undertaken to create a "benchmarking suite" that is separate from the unit testing suite. The benchmarking suite could incorporate additional data (such as that compiled in the IDST and more data for mixed salt systems. When invoked, perhaps the suite could track a cumulative error metric (such as RMSE) over each property of interest.

Having this suite in place would serve three purposes:

1. streamline the unit testing process for new code by eliminating redundant tests
2. facilitate comparing the accuracy of different modeling engines
3. provide quantitative estimates of the model's accuracy for different purposes