fix: separate audio domain and restore transcription templates

Move speech and transcription internals into the audio domain, restore template-driven transcription providers, and regenerate Swagger/SDK so the frontend can stop hand-calling /transcription-* APIs.

conf/providers/deepgram-transcription.yaml

@@ -0,0 +1,9 @@
+name: Deepgram Transcription
+client_type: deepgram-transcription
+icon: deepgram
+base_url: https://api.deepgram.com
+
+models:
+  - model_id: nova-3
+    name: Nova-3
+    type: transcription

conf/providers/elevenlabs-transcription.yaml

@@ -0,0 +1,9 @@
+name: ElevenLabs Transcription
+client_type: elevenlabs-transcription
+icon: elevenlabs
+base_url: https://api.elevenlabs.io
+
+models:
+  - model_id: scribe_v2
+    name: Scribe v2
+    type: transcription

conf/providers/google-transcription.yaml

@@ -0,0 +1,9 @@
+name: Google Transcription
+client_type: google-transcription
+icon: google-color
+base_url: https://generativelanguage.googleapis.com/v1beta
+
+models:
+  - model_id: gemini-2.5-flash
+    name: Gemini 2.5 Flash
+    type: transcription

conf/providers/openai-transcription.yaml

@@ -0,0 +1,9 @@
+name: OpenAI Transcription
+client_type: openai-transcription
+icon: openai
+base_url: https://api.openai.com/v1
+
+models:
+  - model_id: gpt-4o-mini-transcribe
+    name: GPT-4o Mini Transcribe
+    type: transcription

conf/providers/openrouter-transcription.yaml

@@ -0,0 +1,9 @@
+name: OpenRouter Transcription
+client_type: openrouter-transcription
+icon: openrouter
+base_url: https://openrouter.ai/api/v1
+
+models:
+  - model_id: openai/gpt-4o-mini-transcribe
+    name: OpenRouter Transcription
+    type: transcription