Reloading ingress behind orchard controller may yield `no worker registered with this name`

[ruimarinho]

After nginx is reloaded, some bug is causing the orchard controller to lose a worker reference, rendering commands such as orchard vnc vm <id> and orchard ssh vm <id> unusable.

The only workaround is restarting the orchard worker.

I suspect nginx may be dropping the gRPC stream behind port forwarding, but this probably needs more investigation.