_update_min_padding() is called before the backend is "started".
In GR, this triggers an automatic initialization and starts the
default output driver (X11, Quartz or GDI). This doesn't make
sense in headless testing environments (CI) or when using Plots
in Jupyter notebooks.
These patches correct this behavour - but it's unclear (to me),
why the call of _update_min_padding() can't be postponed.
8beee23e21