This function appears to have been incorrect when comparing tree and non-tree entries.

In order to compute the correct diff during a cherry-pick, we need to dehydrate the merge-base as well. Otherwise, the diff will logically delete a lot of files, which may not exist at the target commit.

The results on the gecko-dev repository (https://github.com/mozilla/gecko-dev) show around 500x speedup for in-memory cherry-picks:

```
$ PATH_TO_REPO="$HOME/workspace/gecko-dev" cargo bench cherrypick
Benchmarking Cherrypick/Repo::cherrypick_commit: Warming up for 3.0000 s
Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 19.0s.
Cherrypick/Repo::cherrypick_commit
                        time:   [1.8935 s 1.9011 s 1.9088 s]
                        change: [-1.7816% +2.5204% +7.0223%] (p = 0.29 > 0.05)
                        No change in performance detected.
Cherrypick/Repo::cherrypick_fast
                        time:   [2.7665 ms 2.8209 ms 2.9297 ms]
                        change: [-1.1128% +1.4992% +4.9619%] (p = 0.38 > 0.05)
                        No change in performance detected.
```

Git itself calls the command `git cherry-pick`, but `libgit2` exposes functions called `cherrypick`. This commit separates the term into two words.