“Went dark” timer for soil probes — reading-based (when the reading stopped moving / went null), independent of the unreliable firmware fault flags

Operator request (2026-09-09): ‘if a sensor goes dark I need it to start a timer of when.’ Context: the pot2/probe2 confusion – pot2’s probe was producing a live moving dryback (8.8->8.7) while the firmware sensor_fault/modbus_online flags were both raised (false fault), so the gauges hid the working reading as ‘no data / PROBE DARK’ while the chart showed the real curve. BUILT: a reading-based dark detector that ignores the unreliable firmware flags and decides from whether the MOISTURE READING is actually moving. settings.last_reading_ts() = newest non-null sample; settings.last_change_ts() = newest sample differing from the current value (two indexed queries). sensor_trust._reading_dark_since(): dark since the reading last moved (frozen probe, basis ‘flatline’, after 45 min) or since the last real reading (current value null, basis ‘no_reading’); None while the reading is changing (alive). Derived from fleet_history so the timer survives a brain restart (an in-memory ‘since’ would reset every deploy). Emits binary_sensor.dsc_probe_reading_dark with attrs dark_since / dark_for_s / basis, mirrored onto pot.values (reading_dark_since/_for_s/_basis). RootPage shows a ‘DARK ‘ chip on the probe header with the since-time + basis in its title. Tests: flatline / null / live / never-moved. Full suite 509 passed; deployed (dsc-hub-brain:8.1.0, SPA index-B4kDHWyP). VERIFIED LIVE: probe1 DARK 22.5h (basis no_reading, since 00:53), probe2 DARK 0.0h (caught the transition to null at 23:25). FOLLOW-UP (operator’s call, not done): the gauges still gate on the firmware fault flag (readingOk in RootPage), so a working-but-misflagged probe like pot2 still shows ‘no data’ – switching that gate to the reading-based signal would show its live numbers.