dsh-dafeiyu-emotion: Equipping DSH with a fish ear that understands tone

This is a voice emotion plugin for DeepSeek Harness (v0.3.2, MIT, runtime dsh >=0.2.1-rc.2 <0.3.0, node ^22.19.0 || >=24.0.0). Author LiDu-china, 2 stars on GitHub, categorized as model inference in the community directory. It adds a "Fish Ear" button next to the input box: performs speech-to-text locally, calculates prosodic features in the browser (speaking rate, pauses, fundamental frequency, pitch fluctuation, energy), and fuses a small Chinese and English emotion vocabulary on the server to generate seven categories of emotion labels, confidence, valence, arousal, and tone summary. It populates the composer with "user's original words + emotional context" and sends it automatically, instructing the model to naturally continue the emotion without reading minds. The lightweight mode relies on DSH's built-in Voice Input / SenseVoice, requiring no Python, GPU, or additional model paths; the enhanced mode points `backendRoot` to an existing VoiceSense project, using SenseVoice for transcription, emotion2vec for acoustic emotion, and finally fusing results via BERT + dictionary + prosody. Installation is via npm / GitHub source / Release tarball. The "Fish" replies are provided by dsh-dafeiyu to define the character, while this plugin handles auditory and emotional aspects. TTS defaults to the Piper local model; if switched to "edge", the reply text is sent to Microsoft. Audio is not uploaded, does not enter Session events, and is stored only in memory and local temporary directories.