feat(voice): auto-detect speech language for transcription (drop forced 'en')

Whisper previously hardcoded language:en, mis-transcribing id/en-mixed
speech. Omitting 'language' makes Whisper auto-detect. Paired with enabling
AI_VOICE_TRANSCRIPTION_ENABLED (BWS secret gmw_ai_voice_transcription_enabled
= true) so new recordings are transcribed.

Spec: .hermes/plans/2026-08-30_recordings-v2-features-spec.md

services/discord-gateway/src/modules/voice-recording/voiceTranscriber.ts

@@ -31,7 +31,7 @@ export async function transcribeRecording(
     const response = await openai.audio.transcriptions.create({
       file: createReadStream(oggPath),
       model: "whisper-1",
-      language: "en",
+      // No `language` param → Whisper auto-detects (handles id/en mixed).
       response_format: "text",
     });