BUG: radio fits notebook is broken

[neutrinoceros]

At least one of the notebooks that uses it in the docs is currently broken for a different reason:
https://yt-project.org/docs/dev/cookbook/fits_radio_cubes.html?highlight=fits%20radio

...
ValueError: Invalid vmin or vmax

Originally posted by @neutrinoceros in #4235 (comment)

[neutrinoceros]

here's a minimal reproducer for the broken part of the notebook:

import yt

ds = yt.load("grs-50-cube.fits.gz")
prj = yt.ProjectionPlot(
    ds, "z", ("fits", "temperature"), origin="native", weight_field=("index", "ones")
)  # "ones" weights each cell by 1
prj.set_log(("fits", "temperature"), True)

prj.save("/tmp/")

The reason why this crashes is that we're forcing a log norm but the data contains strictly negative values, as can be seen by leaving the default scaling:

[neutrinoceros]

A possible fix is to manually set zmin to a strictly positive value first, e.g;

prj.set_zlim(("fits", "temperature"), zmin=(1e-3, "K"))

which yields