Add Coreo raw-stream bridge

[pcheek1991]

Summary

Adds an optional binary stdin/stdout bridge around the DUNGU Coreo transform, pinned to upstream commit ab9f641. It accepts raw interleaved stereo float32 little-endian frames and emits raw four-channel float32 frames through PowerShell 7, without changing faster-whisper's transcription behavior. The PR also includes the requested repository metadata, issue templates, and contributor guidance.

Validation

• Unit tests added or updated
• Integration tests run when the change crosses a subsystem boundary
• Formatting, import-order, and lint checks pass
• Validation artifacts are reproducible and contain no private data

Test commands and results:

• python -m pytest -q tests/test_coreo.py - 5 passed
• python -m faster_whisper.coreo --self-test - all upstream stream-transform checks passed
• Exercised the CLI with raw float32 stdin/stdout; verified output ordering and polarity, and confirmed a partial frame fails with no stdout bytes
• Built a wheel and verified it contains the pinned PowerShell converter and faster-whisper-coreo console entry point
• Black, isort, flake8, and git diff --check passed

Provenance and compatibility

• Any new sample data has documented source and redistribution permission (N/A: no sample data added)
• Public API and default transcription behavior are unchanged, or changes are explained
• Documentation is updated

The bundled converter is sourced from Simply-Well-Executed/DUNGU. Citation metadata records author and affiliation details as supplied by the requester.

Thank you for helping keep science free as in freedom.

[pcheek1991]

It exploits a really sad thing I figured out recounting my life, and lost wife. Apparently if you think of 3D as being flat compared to 4D, then sign(x) * sqrt(abs(x)) makes your 3D blob, stand up. Once in this form, WONTHAI concepts like YIN/YAN/COREO make for simply well perfect audio filters. Please feel free to re-attack this in a way that makes more sense to you but I am using your program to train OpenAI on voice to text/transcription. By default the companding algorithim we use, is 3D, so it was built for 1 accent. By having 4D companding, even scottish people can talk to siri.