Detect/correct invalid HP/LP settings in EDF/BDF/GDF

[cbrnr]

Fixes #8582. EDF/BDF/GDF files can contain invalid LP/HP filter settings (cutoff frequencies for HP can be greater than the one for LP). This PR (1) issues a warning and (2) sets HP/LP values for affected channels to None.

[cbrnr]

@apospraf loading files from BCI Competition IV 2B works as expected with this PR.

[cbrnr]

NumPy installation fails in one job?

[larsoner]

Want to add a quick test via monkeypatch (probably of _read_edf_header)? If not I can see if it's easy and push if so

[cbrnr]

Please go ahead and add a test, I won't have time until tomorrow.

[larsoner]

I don't think the current approach quite works properly, for example it doesn't handle '0.0' and '0.000' because they are strings

>>> '0.0' < '0.000'
True

Moreover sometimes edf_info['highpass'] can have length zero and edf_info['lowpass'] can have length n_channels, which fails the NumPy < comparison on latest NumPy dev. It seems safer to deal with these being reversed when actually setting info['highpass'] and info['lowpass'], but maybe not. I'm pushing a commit with this approach and a test that asserts it works for the monkeypatch code at least.

[agramfort]

thx @cbrnr

[larsoner]

@cbrnr all good now?

[cbrnr]

Yes, perfect!