Implement real unit testing.

Right now the unit tests just require that an end-to-end analysis run both in "reproducible" mode (i.e., nsample=0 and using the mean expectation value for PDF-based parameters) and "sampling" mode (i.e., nsample>0 and sampling from PDFs).

We should extend this to actually test the outputs.   Also, we should add some other tests to catch functionality not implement in the example analysis.