Fix power scaling in open_source_scada example + add README and sanity test

[kilojoules]

Summary

The observed power timeseries in
examples/cases/open_source_scada/outputs/observedPower.nc previously
contained power values multiplied by the rated-power numerical value in MW
(1.65), producing a hard cap at 2722.5 kW. This is inconsistent with the
Vestas V82 1.65 MW power curve packaged in plant_energy_turbine/,
which has a rated cap at 1650 kW.

This PR makes the dataset directly consistent with the packaged OEM curve
and adds documentation + a regression test.

What changes

1. outputs/observedPower.nc — power values divided by 1.65. Hard cap
   restored to 1650 kW; bin-median agreement with OEM V82 curve within
   1–15% across operational wind speeds (e.g. WS=13 m/s: OEM 1650 kW,
   observed median 1644 kW). Attributes updated with units, source,
   and license.
2. README.md — documents the source farm (Twin Groves, McLean County,
   IL; Invenergy), turbine type, time range, units, conventions, and the
   pre-PR scaling history.
3. tests/sanity_open_source_scada.py — asserts max(power) ≤ rated
   and |bin-median - OEM| / OEM ≤ 0.20 across WS bins 5–13 m/s. Catches
   future regressions of the scaling bug.

Verification

$ python examples/cases/open_source_scada/tests/sanity_open_source_scada.py
Checking outputs/observedPower.nc ...
  max(power) = 1650000.0 W (rated = 1650000.0 W)
  WS= 5- 6: OEM= 226500 W  median= 193389 W  rel-err=0.15 [OK]
  WS= 7- 8: OEM= 634500 W  median= 544428 W  rel-err=0.14 [OK]
  WS=10-11: OEM=1394500 W  median=1297911 W  rel-err=0.07 [OK]
  WS=13-14: OEM=1650000 W  median=1644006 W  rel-err=0.00 [OK]
  NaN fraction (operational gaps): 0.20  (expect ~0.20)
Sanity check passed.

Why this matters

Without the fix, any downstream code that uses the packaged OEM curve to
reason about this case (e.g. power-curve inversion for inflow estimation,
or per-bin error metrics) silently produces nonsensical results because
the observed values are 65% higher than what the curve predicts at every
WS bin. After the fix, the example is directly usable as a real-onshore
SCADA benchmark with a known V82 turbine.

Notes

• The fix is backward-compatible for anyone who had the old file: dividing
  by 1.65 reproduces the new values exactly. The README documents this so
  users with cached copies can recover.
• The rotor_effective_velocity variable name is left unchanged for
  compatibility, but the README clarifies it is the raw nacelle-anemometer
  reading (no NTF correction applied).
• License unchanged (MIT, inherited from the repository).

Test plan

• sanity_open_source_scada.py passes after the fix
• observedPower.nc opens cleanly with xarray
• Bin-median power vs OEM PC matches within 15% across operational WS

[kilojoules]

https://en.wikipedia.org/wiki/Twin_Groves_Wind_Farm