Fix CUDB equilibrium time unit so g-file header time is correct

[kripnerl]

Problem

Equilibria loaded from CUDB (Fiesta) via pleque.io.compass.cudb / get_ds_from_cudb carry their time axis in seconds. However, Equilibrium.__init__ defaults time_unit to "ms" whenever the input dataset does not explicitly state it. geqdsk.write then encodes the header time as int(time) seconds interpreted as milliseconds:

if equilibrium.time_unit == 's':
    time = int(equilibrium.time * 1000)
else:
    time = int(equilibrium.time)   # <-- 0.65 s written as 0 ms

So writing a GEQDSK file for e.g. COMPASS-U shot 11600 at t = 0.65 s produced a header stamped 0ms instead of 650ms.

Fix

• pleque/io/compass.py — get_ds_from_cudb now records time_unit = 's' on the dataset, symmetric with the existing dst['shot'] = int(shot) handling. This makes CUDB-loaded equilibria self-describing.
• pleque/core/equilibrium.py — time_unit read from a dataset is now normalised to a plain str (it was previously stored as a raw 0-d xarray.DataArray).
• tests/test_gfile.py — added test_time_unit_from_dataset_seconds, a regression test built on the existing test00.nc fixture (whose time is in seconds), asserting time_unit == 's' (a str) and that the GEQDSK header carries the correct millisecond value.

Verification

pytest tests/test_gfile.py tests/test_equilibria.py tests/test_omas.py tests/test_cocos.py tests/test_io.py → 107 passed, no regressions.

🤖 Generated with Claude Code