test(opam_solver): solving sudokus

[art-w]

"Because it's there."

[Alizter]

While this is a cute example, I don't think we can afford having a monolithic test that takes 3s to run, especially if it isn't testing something directly. This test is doing some broad coverage and also acting as a benchmark. I think we should split this.

I would recommend:

1. Extending conflict-class.t to have a longer conflict class list perhaps with another package that shares an entry too.
2. Improving the observed trace in portable-lockdirs-no-solution.t so we can get a better failing observation. This should be able to see opam file reads are cached.
3. Checking sat-solve-trace.t doesn't do any decisions or conflicts.

This should be all the coverage given by this test without having to endure the slow opam parsing.