Zero end-to-end operator automation/rule-authoring UI exists

DONE (v1) in PR #183. A flat trigger -> action rule engine on Settings -> Brain -> Automations. Rule = {trigger:{entity_id,op,value}, action:{type,params}}; ops gt/lt/gte/lte/eq/ne/is/is_not; actions banner and oos_seat only — NO relay/setpoint writes in v1 (deliberate safety boundary; those stay with the climate loop / Zigbee task binding). Disabled by default, edge-triggered with per-rule owned-state, fails closed (hub offline / missing entity / non-numeric => not firing + clears). Evaluator runs in the /ws/fleet loop + /fleet GET, wrapped so a rule bug can’t kill either. 6 brain tests (test_automation_rules.py), full suite 259 passed. Follow-ups: relay/setpoint actions, compound triggers (AND/OR) + time windows + debounce, triggers on computed entities, per-entity staleness horizon.