This release fixes critical data corruption issues that affected both consumption and cost sensors in previous versions.

Fixed Issues

1. Race condition in historical fetch (sensor.py)

   • async_add_external_statistics() is non-blocking, causing offset to be
     read before previous month's data was written to DB
   • Fixed by reading offset ONCE at start and updating mathematically
   • Each month now receives the previous month's final offset directly

2. Cost sensor using wrong values (sensor.py)

   • Cost was calculated using point['cumulative_kwh'] (without offset)
   • This caused costs to reset to near-zero each month
   • Fixed by using final_value (which includes offset) instead

3. Thread safety in migration (init.py)

   • Migration tried to call clear_statistics() from main thread
   • Recorder operations must run in recorder thread
   • Fixed by deferring statistics clearing to background task

Migration (v3 → v4)

Existing users (v1.5.0 and earlier) will automatically:

• Clear ALL corrupted statistics (both consumption and cost)
• Re-fetch complete historical data with fixes applied
• Rebuild database with correct cumulative values

New users install directly with v4 (no migration needed).

Testing

Verified in production with:

• 14,015 data points spanning 20 months (584 days)
• 99.993% accuracy (1 missing point is DST anomaly: 2025-03-30 02:00 doesn't exist)
• All consumption values monotonically increasing
• All cost values correctly calculated with offset

Fixes #10
See: #10 (comment)