v0.3.1

Fixed

• Phantom kWh spike in the cumulative sum at every local midnight. The
  trailing rewindow looked up its cumulative-sum baseline at fetch_start
  UTC midnight, but AGL's period= query is interpreted in the contract's
  local timezone — the first interval of a day query lands at
  (fetch_start - 1)T14:00Z for an AEST account. The baseline therefore folded
  ~10 h of about-to-be-overwritten old sums in, and the new chain re-added
  those hours' deltas, producing a phantom +N kWh jump in the recorder sum
  column every local-midnight UTC row. The per-hour state was always correct,
  but the Energy dashboard plots hourly deltas as sum[h] - sum[h-1], so the
  spike was visible there. _import_intervals now resolves the baseline — for
  the aggregate and every per-tariff ToU series — at the actual earliest
  fetched-interval hour, which is correct regardless of timezone or DST.
  Confirmed against the live recorder: 8 phantom spikes at T14:00Z of
  10–27 kWh, including in the ToU consumption_normal series. Regression test:
  test_baseline_looked_up_at_earliest_fetched_hour.
• Recovery for existing installs. New rows inside the current 7-day
  rewindow are rewritten with correct sums on the next poll. Older rows outside
  the rewindow keep their historical phantom — to clear them, delete the
  haggle:* rows from statistics / statistics_short_term and let the
  30-day backfill rebuild from scratch (same procedure as the v0.2.1 undercount
  recovery).

---