Adding thread completion hooks after the network is running should return an error

The documentation for `fdb_add_network_thread_completion_hook` says that it must be called prior to running the network, but it doesn't enforce this. Adding hooks after the network is running could lead to races with firing the hooks, which could result in unpredictable behavior. Instead, we should detect and fail to add the completion hook once the network is running.

Start by locating fdb_add_network_thread_completion_hook and the code that tracks whether the network is running. Review the function's documentation and lifecycle checks, then verify that adding a hook after startup reports an error rather than succeeding; add or update coverage if the surrounding tests provide a suitable entry point.

Written by the indexing model from the issue text.