fix(costs): clear _last_soc_pct on midnight reset; wire reset_daily_accumulators (#899)

[jackmcintyre]

Summary

Two related issues:

1. CostTracker.reset_daily_accumulators was dead code (zero callers in custom_components/ — only tests exercised it). The midnight-reset path (tick_scheduler.handle_midnight_reset) inlined the same field-zeroing logic but did NOT clear _last_soc_pct.

2. The omission corrupts grid_charge_efficiency: a grid charge spanning midnight leaves _last_soc_pct at the pre-midnight SOC. The first accumulate_costs after midnight computes soc_delta = current_soc - _last_soc_pct and attributes the entire overnight SOC gain to the new day's soc_gain_during_grid_charge_kwh_today, inflating the ratio.

Changes

• utils/costs.py:reset_daily_accumulators: now also clears self._last_soc_pct = None so the first post-midnight grid-charge sample establishes a fresh baseline.
• coordinator/tick_scheduler.py:handle_midnight_reset: delegates to cost_tracker.reset_daily_accumulators(data) (with a defensive inline fallback if cost_tracker is somehow None). Replaces the inline field-zeroing that was duplicating CostTracker's logic and missing the SOC baseline clear.

Validation

• TDD: 2 failing tests in TestAccumulateEnergyKwh:
  • test_reset_daily_accumulators_clears_last_soc_pct — _last_soc_pct must be None after reset (was 40.0).
  • test_no_phantom_soc_gain_after_midnight_reset — end-to-end: pre-midnight charge to SOC 50, reset, post-midnight sample at 81 must NOT attribute the 31-point delta (was 50.0, not None).
• All 57 CostTracker + tick_scheduler tests pass.
• Full suite: 3216 passed, 1 xfailed. Coverage: 95%.
• uv run ruff check clean.