- Fix SmartPI diagnostic sensor recreation after VT reload
- Restrict SmartPI A learning fallback to calibration. Require robust ON slope for SmartPI A learning
- Fix restored SmartPI AB confidence state
- Add HVAC and target context to SmartPI learning windows, and skip B learning in HEAT when the passive heat-loss gradient is too weak or outdoor temperature is too close to target.
Full Changelog: 0.3.5...0.3.6