Try rerunning the solver with --reorder-goals when it reaches the backjump limit.

This idea is from https://github.com/haskell/cabal/issues/1780#issuecomment-41036239. It is a way to find solutions to more dependency problems in a reasonable amount of time without enabling `--reorder-goals` by default. `--reorder-goals` sometimes finds solutions to complex dependency problems more quickly, though it usually slows down simple cases.

cabal could first run the solver without `--reorder-goals`, but with a backjump limit. If the first run reached the backjump limit, it could run the solver again, with `--reorder-goals` and a similar backjump limit. This strategy could avoid slowing down the solver in easy cases where the first run finds a solution.