[Don-Yin]

I got [5, 15, 15, 25] from these descending uint64 bounds, rather than [35, 25, 15, 5]:

import cf_xarray
import numpy as np
import xarray as xr

bounds = xr.DataArray(
    np.array([[25, 35], [15, 25], [5, 15]], dtype="uint64"),
    dims=("x", "bounds"),
    coords={"x": np.array([30, 20, 10], dtype="uint64")},
)
cf_xarray.bounds_to_vertices(bounds, "bounds").values

I replaced the two difference-based order checks with adjacent comparisons, so unsigned wraparound and large signed differences cannot flip the order. This also removes the timedelta-to-float conversions: datetime and cftime values compare directly, without changing the output dtype.

I ran pytest -n auto --cov=./ --cov-report=xml, the CI mypy command, pre-commit run --all-files, and python -m build; I also checked the helpers with NumPy 1.26.4 and xarray 2024.7.0.