Maxwell filter reconstruction

Right now maxwell_filter always reconstructs bad channels and then removes their bad label.

I think there's going to be a higher prevalence of bad channels for OPM than SQUID (we're often missing a few). I've had some issues with interpolation, especially for channels near the edge of the helmet (see example; problem channel in top-right was not recording during the session and labelled bad. In generally, missing channels are reconstructed to have high broadband power).

There are some SSS methods that might be better suited for OPM (eg AMM or iterative SSS). But I wonder if it would make sense to make bads resetting optional. For example, could easily add an argument that controls whether _reset_meg_bads is called, interpolating the bad channels but keeping them labelled bad (which would allow the user to change their mind later, I guess).

HFC is another option for OPM for OPM, but it would be nice to support something a little stronger for systems with higher channel counts (eg I find that SSS has better noise suppression).

So: is SSS reconstruction a plausible issue for modalities that are frequently missing channels and does it make sense to give users control over whether bad channels are reset after SSS?

Any other recommendations here would be appreciated as well, even if user control over bads resetting makes sense. Maybe I should be dropping these channels instead?

Start by reading mne/preprocessing/maxwell.py around _reset_meg_bads and the maxwell_filter entry point. Review the issue's OPM reconstruction concerns and linked SSS references before deciding whether an optional bad-channel reset is appropriate. Done should define the supported behavior for reconstructed channels and their bad labels.