ENH: Multi-channel Wiener filter (MWF) native denoiser

[snesmaeili]

Summary

Rebases PR #51 onto current main and hardens the new multi-channel Wiener filter around an explicit, evidence-aware contract.

• Adds canonical MultichannelWienerFilter; MWF remains a documented short alias.
• Implements the zero-delay generalized-eigendecomposition (GEVD) MWF equations from Somers, Francart & Bertrand (2018), including the reference poseig-style positive-rank default, full/integer rank controls, artifact weighting (mu), and relative covariance loading.
• Requires an explicit 0/1/NaN artifact mask or an independent clean reference. The high-frequency detector is available only through an explicit mask_strategy=hf_power opt-in and is documented as an unvalidated mask-authoring heuristic, not part of core MWF.
• Keeps fitting and evaluation separate: fit() freezes the learned spatial operator and transform() never re-estimates a mask or covariance on evaluation data.
• Preserves MNE object subtype and metadata, aligns channels by name, keeps unselected channels unchanged, supports epoch-shaped masks, and checks sampling-rate compatibility for independent references.
• Exposes sample counts, covariance ranks, GEVD values, selected components, regularization loading, and identity/no-op information as fit diagnostics.

Evidence boundary

This implementation follows the authors' public MATLAB equations and locks internal invariants such as full-rank covariance-ratio equivalence and common-unit rescaling. It does not yet claim external numerical parity with the MATLAB toolbox or validated performance for a particular acquisition regime. The authors' delay-embedded extension is also not included; this PR implements the zero-delay method only.

Primary references:

• Somers, Francart & Bertrand (2018), Journal of Neural Engineering, 15(3), 036007.
• Authors' implementation: https://github.com/exporl/mwf-artifact-removal

Tests and documentation

• 30 focused MWF tests pass on Python 3.10, covering explicit asset contracts, GEVD/full-rank invariants, attenuation, rank deficiency, NaN policies, frozen-operator behavior, unit rescaling, and Raw/Epochs/Evoked metadata preservation.
• Ruff check and format checks pass.
• The complete pre-commit suite passes.
• Strict Sphinx (-W --keep-going) passes with gallery execution disabled for the local documentation check.

The full cross-platform/package matrix remains delegated to this PR's GitHub Actions checks.

[codecov]

Codecov Report

❌ Patch coverage is 96.77419% with 9 lines in your changes missing coverage. Please review.
✅ Project coverage is 96.11%. Comparing base (e84d8d0) to head (c05db1a).

Files with missing lines	Patch %	Lines
mne_denoise/mwf/core.py	96.73%	5 Missing and 4 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #51      +/-   ##
==========================================
+ Coverage   96.08%   96.11%   +0.03%     
==========================================
  Files          65       67       +2     
  Lines        8170     8448     +278     
  Branches     1420     1473      +53     
==========================================
+ Hits         7850     8120     +270     
- Misses        152      157       +5     
- Partials      168      171       +3     

Flag	Coverage Δ
unittests	96.11% <96.77%> (+0.03%)	⬆️

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

Files with missing lines	Coverage Δ
mne_denoise/__init__.py	100.00% <100.00%> (ø)
mne_denoise/dss/denoisers/temporal.py	100.00% <ø> (ø)
mne_denoise/mwf/__init__.py	100.00% <100.00%> (ø)
mne_denoise/mwf/core.py	96.73% <96.73%> (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.