Distinguish between "ok" errors and "fatal" errors

For example, merge conflicts and unapproved PR errors should be a different class of errors than cannot-fast-forward-master, since merge conflicts (and related) occur in normal bot operation (the bot is working as intended), whereas not being able to push to master means that the github repo is in a bad state

Also, we should be consistent about the use of `error`. Maybe convert all `error` statements into a fatal error?

No files, tests, or entry points are named. Start by locating the error statements and the handling paths for merge conflicts, unapproved PRs, and cannot-fast-forward-master; document the intended distinction and verify that normal bot-operation errors and repository-state failures are classified consistently.

Written by the indexing model from the issue text.