Ensure all reference channels are used during final average referencing

[a-hurst]

PR Description

Closes #90. This PR fixes the issue by making reference_channels a local variable within Reference.robust_reference() so that it doesn't affect the self.reference_channels variable that's used to select reference channels during Reference.perform_reference().

As an aside, is there any situation where it would make sense to only use a subset of all channels for final average reference calculation, post-interpolation? Wondering if the whole reference_channels parameter is a carry-over from MatPREP that only exists because EEGLAB doesn't have as nice a method of marking non-EEG channels as MNE. In MatPREP, channels are excluded from being sources for interpolation if they're not specified in the "reference channels" parameter, whereas PyPREP just uses all non-bad EEG channels, but I'm not sure if that would ever actually be a problem.

Merge Checklist

• the PR has been reviewed and all comments are resolved
• all CI checks pass
• (if applicable): the PR description includes the phrase closes #<issue-number> to automatically close an issue
• (if applicable): bug fixes, new features, or API changes are documented in whats_new.rst

[codecov-commenter]

Codecov Report

Merging #92 (9a2ea4a) into master (6abe0bc) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master      #92   +/-   ##
=======================================
  Coverage   98.51%   98.51%           
=======================================
  Files           7        7           
  Lines         675      675           
=======================================
  Hits          665      665           
  Misses         10       10           

Impacted Files	Coverage Δ
pyprep/reference.py	96.55% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6abe0bc...9a2ea4a. Read the comment docs.

[a-hurst]

Also, check out the difference this makes in the MATLAB difference plot for the full PREP example in the docs. That example's not even running in MATLAB-strict mode, either!

Before this PR:

After:

[sappelhoff]

is there any situation where it would make sense to only use a subset of all channels for final average reference calculation, post-interpolation?

I don't think so 🤔 you wouldn't even be able to call it an average reference if not all channels that are part of the data are averaged with respect to all of them.

This fix seems good, thanks for catching it!

[a-hurst]

I don't think so 🤔 you wouldn't even be able to call it an average reference if not all channels that are part of the data are averaged with respect to all of them.

That's what I thought. Wondering if that part of the default API (i.e., needing to specify "referenced" and "rereferenced" channels manually in the params dict instead of just defaulting to all EEG channels) might be worth changing before PyPREP officially leaves its "experimental" stage.