Exact-solve for ground state in MockDWaveSampler on small problems

[randomir]

By default, problems of size 16 and smaller will be sampled with greedy sampler and then an ExactSolver-determined ground state will be injected.

Implements a suggestion from #474 (comment).

Close #474.

[jackraymond]

Perhaps docstring could be improved, since it is not clear how sampleset are impacted. Perhaps the docstring for the sample function would be the appropriate place.
Functionally looks good, I wonder if it would be a good idea to set an explicit upper bound on exact_solver_cutoff to prevent users being surprised by slow regimes or ExactSolver errors.

[randomir]

Thanks @jackraymond, I updated the docstring as suggested, except that I kept a reference to ExactSolver (IMO, it's important for users to understand why and how slow this can get), and I did not impose an upper bound on the cutoff (the default is fine, and if someone decides to modify it, they should read the docs and understand the consequences - which are now much clearer, I think, thanks to your suggestions).