force_reload=True on a single dataset makes minitree twice

[JelleAalbers]

We must first make the minitrees from the first dataset given to hax.minitrees.load, since the dask multiprocessing only works if it knows which variables to expect. Since these both get the force_reload option, using a single dataset with minitrees.load and force_reload causes a double remake.

Not a big issue but hints at a bigger problem: having to make the minitrees to know which variables it will produce. This is inconvenient for other reasons too (see e.g. #47).