Add Nemotron Speech ASR model

[jay-tau]

Summary

• add Nemotron Speech Streaming EN 0.6B to the local Parakeet model registry
• bundle and route sherpa-onnx online websocket server support alongside the offline server
• parse sherpa online websocket results while keeping OpenWhispr's existing record-then-transcribe flow

Note: this does not add live/real-time streaming UX. Nemotron's sherpa artifact uses the online websocket protocol internally; OpenWhispr still records audio, then transcribes it.

Tests

• node --check scripts/download-sherpa-onnx.js src/helpers/parakeetWsServer.js src/helpers/parakeetServer.js src/helpers/parakeet.js src/helpers/parakeetWsResult.js
• node --test test/helpers/parakeetModelRegistry.test.js test/helpers/parakeetWsResult.test.js
• git diff --check
• git diff --cached --check

Closes #986