Follow-up from "refactor: always update on LS state changes"

The following discussion from !1832 (merged) should be addressed:

• @markrian started a discussion: (+1 comment)

  note: I'm starting to think we should be passing state rather than engaged in the event emitter. This would allow us to use the diffEmitter abstraction again.

  In fact, I wonder if engaged is a useful property at all, given it's completely derivable from state 🤔

  That's a larger refactor, though, and outside the scope of this MR.

Also related: gitlab-org/editor-extensions/gitlab-lsp!638 (merged)