Fix tfr db

[drammock]

closes #11091

after reading through the TFR code, to me it seems that #10978 introduced a mistake. Specifically, the changed coef here:

mne-python/mne/time_frequency/tfr.py

Lines 2439 to 2440 in 7127158

	if dB:
	data = 20 * np.log10(data)

is only appropriate for complex data (which have been converted to real-valued amplitude a few lines above). Data that is already real-valued will be power, not amplitude, and so should have coef 10 not 20 in the dB conversion.

My only hesitation is the case of avg_power_itc where the data will be complex, but the complex part is ITC, so it seems like the test of if np.iscomplexobj is too coarse here:

mne-python/mne/time_frequency/tfr.py

Lines 2413 to 2414 in 7127158

	if np.iscomplexobj(data):
	data = np.sqrt((data * data.conj()).real)

...and data in the form avg_power_itc needs to be special-cased to simply discard the imaginary part and get treated like the (non-complex) power cases. @alexrockhill @agramfort @larsoner since you three authored/reviewed #10978 can you confirm that I'm interpreting the code correctly here and that what I've said makes sense?

[drammock]

data in the form avg_power_itc needs to be special-cased to simply discard the imaginary part and get treated like the (non-complex) power cases.

I was wrong about this. They're only combined as power + 1j * ITC internally and temporarily. By the time we're plotting, the real & imag parts have been split into separate (real-valued) TFR objects.

(so the np.iscomplexobj test is sufficient)

[larsoner]

compat/old is unhappy:

_____________________________ test_dB_computation ______________________________
mne/time_frequency/tests/test_tfr.py:574: in test_dB_computation
    quadmesh1 = fig1.axes[0].collections[0]._mapped_colors
E   AttributeError: 'QuadMesh' object has no attribute '_mapped_colors'

maybe just wrap with if hasattr ?

[larsoner]

Thanks in advance @drammock , marking for auto merge