feat(dss): native MNE covariance support for Evoked inputs

[snesmaeili]

Summary

DSS._fit_mne fell back to the internal NumPy covariance path for Evoked
inputs, because mne.compute_covariance accepts only Epochs and Raw. This
adds a small compute_evoked_covariance helper that treats the averaged
response's time samples as observations and returns a genuine mne.Covariance,
then wires it into the Evoked branch of _fit_mne so the estimation stays
within the MNE ecosystem — consistent with the Raw and Epochs branches.

Changes

• mne_denoise/dss/utils/covariance.py: new compute_evoked_covariance. It
  wraps the Evoked as a single-trial EpochsArray and defers to
  mne.compute_covariance with keep_sample_mean=True (so the single average
  is not subtracted away). The intentional "not baseline corrected" notice is
  suppressed, since DSS uses the raw second-moment covariance.
• mne_denoise/dss/linear.py: the Evoked branch builds baseline/biased
  covariances through the helper instead of calling _fit_numpy.
• Tests: unit tests for the helper (type/shape, second-moment value, guard on
  too-few samples) plus a regression test asserting the Evoked fit reproduces
  compute_dss on the MNE-side covariances.

Closes #39

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.10%. Comparing base (e84d8d0) to head (441e596).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #42      +/-   ##
==========================================
+ Coverage   96.08%   96.10%   +0.02%     
==========================================
  Files          65       65              
  Lines        8170     8187      +17     
  Branches     1420     1423       +3     
==========================================
+ Hits         7850     7868      +18     
  Misses        152      152              
+ Partials      168      167       -1     

Flag	Coverage Δ
unittests	96.10% <100.00%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
mne_denoise/dss/linear.py	99.31% <100.00%> (+<0.01%)	⬆️
mne_denoise/dss/utils/__init__.py	100.00% <100.00%> (ø)
mne_denoise/dss/utils/covariance.py	100.00% <100.00%> (ø)

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

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

[BabaSanfour]

Thanks for implementing this sina; but I was mainly thinking whether we could make mne-python mne.compute_covariance function accepts an evoked object too? that would be the cleanest way of doing it. @drammock @larsoner what do you think of this? I can open the issue on mne-python to make the change; if it is out of scope for mne-python we can work on the changes @snesmaeili suggested.

[larsoner]

Yeah I think in principle compute_covariance could accept evoked at the MNE level

[BabaSanfour]

Amazing, I will open a PR and work on it in mne; probably next week. Thanks.

[BabaSanfour]

This PR depends on mne-tools/mne-python#14129 (mne-tools/mne-python#14129), which closes mne-tools/mne-python#14115 (mne-tools/mne-python#14115). It will remain a draft and can be merged once that PR is merged and MNE-Python 1.13 is released, since these changes intentionally require mne>=1.13.