Optimize PrepareRelease performance

[dbanty]

In order to fix #505, we started having to parse every commit in history and keep them in memory in order to select the correct commits. This is probably still fast enough for most projects, but really inefficient for large ones and probably consumes a ton of memory.

So... there are a couple of improvements we could make:

1. Allow folks to completely disable commit-traversal if they're using changesets only
2. Use a graph-aware algorithm to determine if a potential commit's parents are in the history of the previous releases without traversing the entire graph or storing everything in memory
3. As an alternative to 2, start shelling out to git rev-list or something. If we start relying on git to be present (if using conventional commits), then we can also drop libgit2 and probably gix which each make up a massive chunk of the compile time.