Music filter engines for ContextLens

These are third-party models, redistributed unmodified with their licences. ContextLens uses them to take the music out from under a spoken lesson while keeping the voice.

Two licences apply, one per folder:

• kim-mixA/ — Kim's vocals Mel-Band RoFormer, MIT. Converted to a Core ML package with 4-bit mask estimators; the weights are the author's. Three files, which together are the .mlpackage folder as macOS expects it.
• mossformer2-48k/ — MossFormer2 SE 48K (ClearVoice), Apache-2.0, exported to ONNX and quantised to 8-bit.

The app's own low tier, DPDFNet-2 (Apache-2.0), is not here: it ships inside the app.

Each file's size and sha256 are listed in the app's resources/music/engines.json, and a file whose hash does not match is refused rather than installed.