[BUG] OPF `time_elapsed` computed with `.seconds` and `int()` — sub-hourly runs silently decouple storage power from energy

[joda9]

Where

edisgo/io/powermodels_io.py (in to_powermodels):

pm["time_elapsed"] = int(
    (psa_net.snapshots[1] - psa_net.snapshots[0]).seconds / 3600
)  # length of timesteps in hours

This single global value drives all inter-timestep couplings in the Julia OPF (battery,
heat storage, DSM energy balances, EV charging energy).

Three independent defects

1. .seconds instead of .total_seconds() — .seconds is the seconds component of the
   Timedelta (0–86399), not its total duration. A gap of exactly 1 day yields 0; a gap of
   25 h yields 1 h. Silently wrong.
2. int(...) truncation — 15-min resolution: int(900/3600) = int(0.25) = 0. With
   time_elapsed = 0 the storage coupling se₂ − se₁ == −0·ps₂ degenerates to 0 == 0 for
   any ps: power and energy are completely decoupled. The battery becomes an
   inexhaustible storage with no energy accounting (results look plausible — power bounds are
   respected — but are physically meaningless). With flexible charging points the model becomes
   infeasible instead (cpe₂ == cpe₁ is forced while the flexibility-band midpoint moves).
   Real 15-min runs have been executed and were affected by this.
3. Only the first snapshot gap is used, globally — non-equidistant time indices are
   silently mis-scaled for every step after the first.

Suggested fix

• Use total_seconds(), keep the value a float (the Julia side handles non-integer hours;
  note the interaction with the heat-storage standing-loss exponent, see [BUG] Heat-storage standing loss uses fixed exponent (1-p_loss)^(1/24) — wrong for non-hourly timesteps #697).
• Enforce the implicit contract fail-loud in to_powermodels: snapshots must be
  equidistant and contiguous, otherwise raise. The OPF's time-coupling formulation is
  meaningless across gaps.