If the EV's service rejected a write (e.g. Tesla integration raising
'Command was unsuccessful: is_charging' when we redundantly call
switch.turn_on while charging already running), the exception used to
propagate up through _async_update_data, killing the whole coordinator
update and putting the integration into setup_retry indefinitely.
Wrap service calls in a _safe_service_call helper that catches and
logs at WARNING; the coordinator still returns a valid Decision so
entities populate. The next tick reconciles naturally.