Improved Trace Generation

[DavePearce]

At the moment, trace generation adopts a "brute-force" approach. There are some issues with this:

1. (Persistent Failures). A trace can fail due to a problem early on in the table. Thus, any other trace which contains this pattern will fail for the same reason. In particular it doesn't seem worth creating traces which are larger than this trace. Note, however, that some constraint forms do have non-monotonic behaviour (e.g. lookups).

To address (1) above, we could adopt an incremental approach. That is, we grow the traces from 0 lines upwards. Failing traces are then culled earlier and do not lead to a large number of traces failing for the same reason.

Another solution to the above is via sampling.