[REQ] BS-RoFormer separation model(s)

Hi there,
I've recently tested the model (which seems to outperform the best Demucs version (htdemucs), according to [this MVSEP's - vocal - leaderboard](https://mvsep.com/quality_checker/multisong_leaderboard?sort=vocals) (about 3 sdr points: 11.8938 > 8.7869).

My (lossless) demos:
- https://mvsep.com/result/20260318182458-252cd55a1d-un-cavalier-di-spagna.flac
- https://mvsep.com/result/20260318193400-a786943fd2-le-corsaire.flac

It would be interesting to exploit this so here's some - maybe - useful links:

HF: https://huggingface.co/models?sort=modified&search=BSRoformer

ONNX: https://huggingface.co/xycld/BS-RoFormer-ONNX

CPP version by @chenmozhijin: [BSRoformer.cpp](https://github.com/chenmozhijin/BSRoformer.cpp)

Thanks in advance.

The issue links BS-RoFormer model listings, an ONNX export, and a C++ implementation, but names no repository files, tests, or entry point. Start by locating the repository's audio-separation integration and comparing it with the linked ONNX and C++ implementations; done requires an agreed integration scope and working BS-RoFormer separation support.

Written by the indexing model from the issue text.