Reuse ActorContainerRef if it exists

[MellowYarker]

Prior to this commit, if a Durable Object in Workerd (local dev/miniflare) hibernated (i.e. Worker::Actor was destroyed), but the stub wasn't dropped by the worker (i.e. PromisedWorkerInterface still alive), then we would incorrectly create an entirely new refcounted ActorContainerRef upon a new request coming in.

This is incorrect because the old ActorContainerRef is still owned, and by creating an entirely separate one, we break our request tracking.

Specifically, the ActorContainer only tracks a weak reference to an ActorContainerRef, and so if we have two separate ActorContainerRef's one of them will not be tracked.

[MellowYarker]

I suspect this may fix #1422, although I haven't been able to force a segfault in server-test.c++ yet. At the very least, it's clear the code prior to this commit is dropping a live ActorContainerRef, and if the ActorContainer goes away before the untracked ActorContainerRef, then the untracked ActorContainerRef will attempt to access invalid memory in its dtor (which is what we see happening in that issue).