Support bisecting from good commits to bad commits

`--bisect` is very helpful for finding when a regression happened, but it doesn't seem to support bisecting from a bad commit to a good commit, i.e. to verify when a bug got fixed. This can be useful to understand when something in Bazel got fixed, and verify it was deliberate.

Start by tracing Bazelisk's existing `--bisect` handling and how it determines good and bad commits. Confirm the expected behavior for a bad-to-good range, then exercise the bisect workflow and verify that it identifies the commit where the bug was fixed.

Written by the indexing model from the issue text.