fix: always persist calibration after sampling, not only when correction moves

[bvweerd]

Root cause

When the efficiency correction hits CALIBRATION_APPLY_MAX (1.05) — which happens when the battery consistently outperforms its configured curve — every new sample is recorded in the in-memory rolling window but record() returns moved=False. Because async_save() was only called when moved=True, the .storage file was never updated.

On every HA restart the stale file (still in the old raw_ratio schema) was loaded, _migrate_ratios() ran again, and the correction snapped back to 1.05. New samples kept coming in and hitting the cap, the file was never written, and the sensor value never changed — until a manual reset via the service call.

Fix

Save after every successful sample (CALIBRATION_SAMPLED), not only when the correction moves by more than CALIBRATION_SIGNIFICANT_CHANGE. This ensures:

• The file is converted to the efficiency_factor schema on the first sample after startup, so subsequent restarts load clean data
• The sample count in the file stays current
• The sensor reflects actual sampling activity even when correction is at the cap

Test plan

• python -m pytest tests/ -v — all 1052 tests pass
• Deploy and verify last_result shows sampled after a charge/discharge step instead of staying at direction_not_planned
• Verify .storage file is updated after each run (schema becomes efficiency_factor, sample count increments)