ovos-ww-plugin-local-wake

The detection method and the feature extraction are adapted from local-wake
(https://github.com/st-matskevich/local-wake), Copyright (c) 2025
st-matskevich, MIT License. Changes:
- The comparison runs on audio chunks from the OVOS listener.
- The DTW is computed with numpy instead of librosa.
- Recordings are trimmed of silence when they load, and a recording
  shorter than the model window is widened with its own surrounding audio.
- A buffer much quieter than the recordings is not compared.
- The buffer length follows the longest recording.

models/speech-embedding.onnx is Google's speech-embedding model
(https://www.kaggle.com/models/google/speech-embedding), Apache License 2.0,
in the ONNX conversion shipped by local-wake.
