feat(channel): add customizable JS model-call scripts with split-panel editor

- Run user-authored async scripts per model with flat locals (prompt,
  images, messages, params, model, baseUrl, apiKey, http, request, poll,
  sleep, signal, onDelta); empty script falls back to system default.
- Wire image/video/audio/text request entrypoints through the plugin
  runtime, keeping default handlers unchanged.
- Split-panel script editor: variable/return docs on the left, CodeMirror
  JS editor on the right; title shows capability - model.
- Provide OpenAI and Gemini template buttons per capability; text template
  uses the /responses API; image template branches text2img vs img2img.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

web/src/services/api/audio.ts

@@ -32,7 +32,8 @@ export async function requestAudioGeneration(config: AiConfig, prompt: string, o
                 capability: "audio",
                 script,
                 config: requestConfig,
-                input: { prompt, params: { voice: normalizeAudioVoiceValue(config.audioVoice), format, speed: normalizeAudioSpeedValue(config.audioSpeed), instructions: config.audioInstructions.trim() } },
+                prompt,
+                params: { voice: normalizeAudioVoiceValue(config.audioVoice), format, speed: normalizeAudioSpeedValue(config.audioSpeed), instructions: config.audioInstructions.trim() },
                 signal: options?.signal,
             });
             return await audioPluginBlob(result, format);