Try and fail if there are multiple workers running with the same worker name

This issue has been migrated from [#9345](https://github.com/matrix-org/synapse/issues/9345).

---

A bunch of stuff assumes that worker names are unique, especially workers that write to streams (and most workers write to the caches stream).

One possible way of doing this is to use postgres application locks, so that we can check every time a postgres connection is created that the running instance is only one with the given worker name. (This is slightly complicated as there will be multiple DB connections from the same process at the same time)