Consider removing the "count conflicts" solver feature, since it may be redundant with "fine grained conflicts"

The "count conflicts" feature significantly improved solver performance when it was implemented, but the "fine grained conflicts" feature was added later and may provide the same benefits. This issue is for running the benchmarks in https://github.com/haskell/cabal/tree/master/solver-benchmarks to compare enabling and disabling `--count-conflicts` while `--fine-grained-conflicts` is enabled. If the average run time is similar, and there are no cases that are handled much better by `--count-conflicts`, then I think we can remove the feature in order to simplify the code.

PRs that added the two features:

#3513 `--count-conflicts`
#5918 `--fine-grained-conflicts`