`generate-time-averages` answers with `cdo` / `numpy=2.*` different from `cdo` / `numpy=1.26.*`

[ilaflott]

saw this in pipeline(s) for #707 , which may or may not have shown other issues that may crop up because of a potential future transition to numpy=2.*