feat: add OpenClaw realtime transcription plugin

[LauraGPT]

Summary

• add the external openclaw-funasr realtime transcription plugin under integrations/openclaw
• support FunASR online, offline, and 2pass WebSocket modes for OpenClaw Talk and Voice Call
• convert OpenClaw 8 kHz G.711 mu-law audio to 16 kHz PCM with official 60 ms FunASR framing
• document configuration, package validation, and the upstream OpenClaw SDK dependency
• add discoverable English and Chinese README entry points

Runtime behavior

The provider negotiates the required binary WebSocket subprotocol, sends the official initialization and end-of-input messages, maps partial/final transcripts, supports hotwords and ITN, and bounds queued audio and retained transcript state. Embedded URL credentials are rejected; optional reverse-proxy authentication uses a Bearer token.

Validation

• npm ci: clean install, 0 vulnerabilities
• OPENCLAW_ROOT=<openclaw-with-#118977> npm test: 14 passed
• OpenClaw external package contract: issues: []
• npm pack: six-file artifact, npm shasum 63ff5b822a3bc47c7187f095b000769682e5a55d
• fresh npm-pack: installation: plugin status loaded, runtime registered realtime-transcription: funasr
• official FunASR server live proof: 278 telephony-format frames, seven partials, exact final transcript, no client errors
• TruffleHog 3.96.0: 0 verified and 0 unverified secrets

Upstream dependency

Publication to npm and ClawHub remains gated on openclaw/openclaw#118977. The compatibility floor will be pinned to the exact OpenClaw release containing that SDK change before publication.

AI assistance

AI assistance was used for implementation and test orchestration. The package contract, tests, packed artifact, runtime installation, and live protocol behavior were verified directly.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.