use better defaults for n_interpolates

[jasmainak]

The current default of n_interpolates=np.array([1, 4, 32]) seems to fail for EEG with smaller channels e.g., 32 or 20. It sometimes finds all channels as bad and tries to interpolate. Found this issue while debugging a recent user problem.

I think a better default would depend on the number of channels. E.g., n_interpolates = np.arange(1, n_channels // 2, n_channels // 8) or something along these lines. Thoughts @agramfort ?

[agramfort]

would this break any existing code to change this? did you evaluate / quantify the impact?

…