Use direct NumPy MAD in the windowed correlation loop

[sappelhoff]

Summary

Replace the two scipy.stats.median_abs_deviation(..., axis=1) calls in the windowed loop of find_bad_by_correlation with the equivalent direct NumPy expression np.median(np.abs(x - np.median(x, axis=1, keepdims=True)), axis=1).

Why

median_abs_deviation carries a scipy axis_nan_policy wrapper (broadcast/concatenate bookkeeping) that adds noticeable overhead when called many times on small arrays — here roughly 2× per correlation window (~1100 windows). On an 18.5-min, 99-channel @ 500 Hz recording:

• the MAD call itself: ~1.58× faster
• find_bad_by_correlation: 5.06 s → 4.51 s (on top of Vectorize bandpass filtering in NoisyChannels._get_filtered_data #195)

Correctness

The default median_abs_deviation (scale 1.0, center = median) is exactly median(|x − median(x)|):

• np.array_equal, maxdiff 0.0 against median_abs_deviation in a direct micro-check.
• Full test suite passes (47 passed), including tests/test_matprep_compare.py.
• get_bads(as_dict=True) and all _extra_info arrays (including the per-window max_correlations, noise_levels, channel_amplitudes) unchanged on the real recording (reject "omit"/None) and the eegbci fixture (matlab_strict False/True).

Note: the same swap would also speed up find_bad_by_nan_flat and find_bad_by_hfnoise; kept out here to keep the diff minimal — happy to follow up if wanted.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.86%. Comparing base (54e2554) to head (5e27631).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #197   +/-   ##
=======================================
  Coverage   97.86%   97.86%           
=======================================
  Files           7        7           
  Lines         842      844    +2     
=======================================
+ Hits          824      826    +2     
  Misses         18       18           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.