NTX Actor Followup

[sergerad]

There are a series of review comments to address after the merge: #1435

---

Also fom comment:

I also think this structure is probably half-way to where we'd need to get to eventually. The main things to address in follow-ups:

1. How we segment the actor state - the current structure does not seem clean to me - but changing it now will probably result in quite a bit of refactoring.
2. How we handle event routing in the coordinator. I think this may have significant performance implications - but also, we should confirm if that's the case first.
3. How to handle actor failures - e.g., under what circumstances we may want to respawn failed actors.

The first point is probably also good to address in combination with introducing lazy account loading as both of these will have impact on how the actor state is structured.