FIX: Fix regression with mne.viz.plot_evoked

[mistraube]

Reference issue (if any)

Fixes #13271.

What does this implement/fix?

Fix a regression with mne.viz.plot_evoked() when using gfp="only" or gfp=True.

Plots based on: https://mne.tools/stable/auto_tutorials/evoked/30_eeg_erp.html#sphx-glr-auto-tutorials-evoked-30-eeg-erp-py

l_aud.plot(gfp="only")

Before:

After:

l_aud.plot(gfp=True, spatial_colors=True, ylim=dict(eeg=[-12, 12]))

Before:

After:

Additional information

I tried to identify any other plotting code that might be negatively affected by this change, but I didn't find any.

[larsoner]

Hmmm in your first plot there is a gap at the bottom, ideally the ylim would start at zero since GFP must be non-negative...

I think in the non-GFP-only case we must do something to set the ylim properly. What is it? Could we extend that logic instead, setting the ylim manually in the GFP-only case? That seems like maybe better/more consistent behavior. Maybe something like ylim = (0, (1.1 * np.max(gfp)) or 1) or similar would work...

[mistraube]

Ok, then actually the behavior before this issue was introduced was already not correct. It also has the gap in the gfp-only case. See: https://mne.tools/1.5/auto_tutorials/evoked/30_eeg_erp.html#sphx-glr-auto-tutorials-evoked-30-eeg-erp-py

I'll look into it again.