With the PR introducing error categories https://github.com/realm/realm-java/pull/6379 the whole error concept is starting to smell quite a bit (a lot).

It roots in the fundamental mismatch between the native layer reporting errors as (string category, int code), while Java tries to make it easy to enumerate errors by forcing them into enum values.

Right now we try to fix this by prefixing the enums with a category description. This might be the best solution but the downside is that the enum descriptions get quite long. Also, the list of error cases are very long which means it becomes overwhelming as it is hard to know which errors to handle and which to ignore.

This PR covers taking a fresh look at how errors are reported and if we can do anything to make it easier, even if it involves breaking changes.