Add ProgramEvidence class to store the results of modular analysis

[ChrisTimperley]

To allow Darjeeling to efficiently exploit the results of multiple, modular analyses, we should add a ProgramEvidence class that simply records named, immutable articles of evidence that can be used during repair (or further analysis steps).

In a subsequent update, we allow individual analysis and repair operations to specify the evidence that they require and produce via requires and produces. We can then use an efficient scheduler to split the workload of collecting evidence across as many cores as possible.

[ChrisTimperley]

This is a more general approach to #265