Fixes #14146

This is an alternative approach to PR #14175

I considered @machine424's initial suggestion to set m.groups to empty slice. However in case Update is called after Close we would end up with new rule groups running and that cannot be stopped using Stop since it is not possible to call Stop twice. Actually the same thing happens with #14175 . The approach in this PR makes for a cleaner shutdown: no dangling go-routines that are stopped in an unpredictable way.

As @roidelapluie mentioned, we don't have any case were we restart a stopped ruler, so I think this approach is acceptable.

Credits to @machine424 for providing the unit tests.

PS:
Another approach that I also considered, but that I don't think it is worth implementing is to make Update after Close return an error, (i.e.: make it illegal), but that would not solve the problem of the race condition during shutdown because it would produce misleading error messages in case there is a reload after SIGTERM.
We would need to either:

• Avoid calling Update during reload in case the manager is already closed by adding a new method to it IsClosed and a mutex that controls the access to the manager.
• Implement error types for the Update method that the reloader could use to distinguish between real errors and expected errors.