Make PUT /instance/state 503 when waiting to init

Also, I noticed that the `PUT /instance/state` API route would return a
`NoInstance` error when trying to change the state returned a
`VmError::WaitingToInitialize`, which seemed potentially bad: this would
result in a sled-agent that tries to send a state change request to a
still-initializing VM to believe it's Permanently Gone, and mark it as
`Failed`, tear down the zone, and so on. Which seems rude of it!

I don't think this is likely to be a problem in practice since IIRC both
sled-agent and Nexus will not try to send state change requests to
instances that they understand to be still initializing, but it seemed
good to not return the INSTANCE IS PERMANENTLY GONE error code here.
Now, we return a 503, so the sled-agent will just know it needs to wait
for a bit.

bin/propolis-server/src/lib/server.rs

@@ -426,7 +426,10 @@ async fn instance_state_put(
         .put_state(requested_state)
         .map(|_| HttpResponseUpdatedNoContent {})
         .map_err(|e| match e {
-            VmError::WaitingToInitialize => not_created_error(),
+            VmError::WaitingToInitialize => HttpError::for_unavail(
+                None,
+                "instance is still initializing".to_string(),
+            ),
             VmError::ForbiddenStateChange(reason) => HttpError::for_status(
                 Some(format!("instance state change not allowed: {}", reason)),
                 hyper::StatusCode::FORBIDDEN,