Extend the SpamCheck API to report more than spam

This issue has been migrated from [#9805](https://github.com/matrix-org/synapse/issues/9805).

---

In the future, we may end up using this API to do more than spam-checking. For instance, if a homeserver is child-friendly, they may wish to have some kind of image filter to detect violent imagery. As this is not spam, it would be nicer if the SpamCheck could return an information such as "against room policy", rather than a simple boolean spam/not spam. This would then let the clients adapt error messages to be more understandable.