Cap time extrapolation gap to prevent charging SOC plateaus

Uncapped extrapolation in get_predicted_soc() inflated prediction
during long MQTT gaps (53 min observed), then the never-decrease
constraint locked in the inflated value. accumulate_energy() already
caps at MAX_ENERGY_GAP_SECONDS (600s) but get_predicted_soc() did
not, creating a mismatch that caused visible prediction plateaus.

Trace: MQTT burst 10:33-10:35, 53-min gap, extrapolation inflated
51.4%->58.7%, real energy stuck at 0.307 kWh, plateau until
re-anchor. After HA restart, 74-min gap from persisted session
inflated to 61.5%, second plateau at 61.7% until API sync-up at 62%.

Author: renaudallard

custom_components/cardata/soc_prediction.py

@@ -974,11 +974,15 @@ def get_predicted_soc(
 
         # Extrapolate energy since last power reading using last known power.
         # This provides smooth SOC updates between sparse API polls.
+        # Cap gap to MAX_ENERGY_GAP_SECONDS to match accumulate_energy() —
+        # without cap, long MQTT gaps inflate prediction, then "never decrease"
+        # constraint locks in the inflated value creating visible plateaus.
         if session.last_power_kw > 0 and session.last_energy_update is not None:
             gap = time.time() - session.last_energy_update
             if gap > 0:
+                capped_gap = min(gap, MAX_ENERGY_GAP_SECONDS)
                 net_power = max(session.last_power_kw - session.last_aux_kw, 0.0)
-                extra_kwh = net_power * (gap / 3600.0) * efficiency
+                extra_kwh = net_power * (capped_gap / 3600.0) * efficiency
                 energy_added_kwh += extra_kwh
 
         # Convert to SOC percentage