Describe the bug

Lowpass and highpass cut off frequencies are swapped when using mne.io.read_raw_gdf() to load a gdf file from the BCI IV 2b dataset.

Steps to reproduce

1. Download BCI IV 2b dataset from here (it requires a name and an email)
2. Load one of the gdf files using mne.io.read_raw_gdf()

Expected results

raw_gdf.info should return:

<Info | 7 non-empty values
 bads: []
 ch_names: EEG:C3, EEG:Cz, EEG:C4, EOG:ch01, EOG:ch02, EOG:ch03
 chs: 6 EEG
 custom_ref_applied: False
 highpass: 0.5 Hz
 lowpass: 100 Hz
 meas_date: 2006-03-07 09:23:36 UTC
 nchan: 6
 projs: []
 sfreq: 250.0 Hz
>

Also raw_gdf.plot() should look like the figure below

Actual results

raw_gdf.info

<Info | 7 non-empty values
 bads: []
 ch_names: EEG:C3, EEG:Cz, EEG:C4, EOG:ch01, EOG:ch02, EOG:ch03
 chs: 6 EEG
 custom_ref_applied: False
 highpass: 100.0 Hz
 lowpass: 0.5 Hz
 meas_date: 2006-03-07 09:23:36 UTC
 nchan: 6
 projs: []
 sfreq: 250.0 Hz
>

raw_gdf.plot()

Additional information

I'm using the 0.22.dev0 version of mne on Ubuntu 18.04