Present CI results in diagnostic categories

[shonfeder]

To help package authors and opam repo maintainers, we should provide clear categorization of the kinds of CI failures that occur.

As per #233 (comment)

Yes, the important thing is regrouping by type of error.
Then you know how many errors there are to repair.

A simple approach would be to just group all identical error messages into the same buckets. More sophisticated treatment would group errors into buckets based on kinds of errors (E.g., missing dependencies, test failures, build failures, etc.)

This issue tracks the work that was sketched out in #233