feat(pvforecast): model snow on the modules in the local backend (#1484)

No PV forecast modelled snow lying on the modules, so after a snowfall the
optimizer planned with PV that never came. The local Open-Meteo/pvlib
backend can now reduce the forecast by the snow coverage (opt-in via
pvforecast.akkudoktor.snow_enabled):

- Request hourly snowfall and snow depth with the weather and run the NREL
  snow coverage model (pvlib.snow) per plane.
- Correct the coverage with measured PV production: production above the
  model clears it at once; low production raises it only while snow is
  plausible, so curtailment and inverter outages are not learned as snow.
  Persisting snow damps the slide rate, fading with
  snow_feedback_memory_hours.
- Rebuild the coverage from snow_lookback_days on every update instead of
  storing state.
- Exclude intervals with possible snow from the calibration fit.
- Warn when PV meter readings are missing or stale; ignore impossible meter
  jumps.

_plane_power keeps its two-value signature (used by the adaptive provider in
#1422); the irradiance comes from the new _plane_power_and_poa.

Tests replay real February 2026 data from one plant (snow on 02-19/20).

Signed-off-by: SchaafAlexander <86607129+SchaafAlexander@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

docs/akkudoktoreos/prediction.md

@@ -908,6 +908,50 @@ Hourly meters are fitted hourly; native quarter-hour meters retain their cadence
 Probable outage or curtailment days are excluded by default. Without sufficient
 usable measurements the uncalibrated physical model is used.
 
+##### Snow on the modules
+
+Set `snow_enabled` to true to model snow lying on the modules. The local backend
+then also requests hourly `snowfall` and `snow_depth` from Open-Meteo and runs the
+NREL snow coverage model (`pvlib.snow.coverage_nrel`) per plane: heavy snowfall
+covers the modules, warmth and sunshine let the snow slide off, and the modules
+count as clear once the ground snow depth drops below `snow_threshold_depth_cm`.
+The forecast is reduced by the covered fraction. No state is stored between runs;
+every update replays the last `snow_lookback_days` of weather to rebuild the
+current coverage.
+
+The weather model alone often misjudges how long snow stays on the modules. With
+`measurement.pv_production_emr_keys` configured (cumulative meter readings in
+kWh), `snow_feedback_enabled` corrects the coverage with the measured production
+of the past hours:
+
+- Production above the snow model clears the coverage at once.
+- Production far below the clean-module model (less than `snow_detection_ratio`)
+  raises the coverage, but only while snow is plausible: snow lies on the ground
+  or the model already holds some coverage. Low production on a snow-free day
+  (curtailment, an inverter fault) is ignored.
+- Persisting snow also slows the modelled slide rate. This fades back with
+  `snow_feedback_memory_hours`, so a warm night after a covered day still clears
+  the modules in the forecast.
+
+Intervals with possible snow are excluded from the calibration fit, so snow is
+never learned as a permanent model loss. Meter jumps larger than the plant can
+produce are ignored.
+
+```json
+{
+  "pvforecast": {
+    "provider": "PVForecastAkkudoktor",
+    "akkudoktor": {
+      "backend": "local",
+      "snow_enabled": true
+    }
+  },
+  "measurement": {
+    "pv_production_emr_keys": ["pv_production_emr"]
+  }
+}
+```
+
 Transient weather errors are retried. If an update still fails, existing usable
 AC forecasts are retained; a cold start or expired-only forecast reports the
 error. Retention does not extend the available forecast horizon. Calibration and