Bad-by-dropout channels sometimes ignored during Reference

[a-hurst]

As mentioned in #93 (specifically this comment), both PREP and PyPREP don't flag bad-by-dropout channels as bad or unusable on the first pass before average referencing. The problem with this is that after average referencing, the dropout regions in any bad-by-dropout channels likely aren't going to be completely flat anymore, so they're unable to be detected unless the bad-by-correlation detector manages to flag them.

To fix this, it seems like initial bad-by-dropout channels should be added to the initial noisy channels state here unless matlab_strict is True:

pyprep/pyprep/reference.py

Lines 219 to 230 in 3625b26

	# Initialize channels to permanently flag as bad during referencing
	noisy = {
	"bad_by_nan": noisy_detector.bad_by_nan,
	"bad_by_flat": noisy_detector.bad_by_flat,
	"bad_by_deviation": [],
	"bad_by_hf_noise": [],
	"bad_by_correlation": [],
	"bad_by_SNR": [],
	"bad_by_dropout": [],
	"bad_by_ransac": [],
	"bad_all": [],
	}

What do you think?