feat(MTMDChatHandler): support audio inputs and fix interleaved media ordering

- Refactored `CHAT_FORMAT` to use a single loop for `message.content`, preserving the exact chronological order of interleaved text, images, and audio.
- Added template routing for `audio_url`.
- Added template routing for OpenAI's `input_audio` format, properly formatting it as a Data URI.

Author: JamePeng

llama_cpp/llama_chat_format.py

@@ -2803,10 +2803,15 @@ class MTMDChatHandler:
                     "{% for content in message.content %}"
                         "{% if content.type == 'image_url' %}"
                             "{{ content.image_url if content.image_url is string else content.image_url.url }}"
-                        "{% endif %}"
-                    "{% endfor %}"
-                    "{% for content in message.content %}"
-                        "{% if content.type == 'text' %}"
+                        "{% elif content.type == 'audio_url' %}"
+                            "{{ content.audio_url if content.audio_url is string else content.audio_url.url }}"
+                        "{% elif content.type == 'input_audio' %}"
+                            "{% if content.input_audio is string %}"
+                                "{{ content.input_audio }}"
+                            "{% else %}"
+                                "data:audio/{{ content.input_audio.format }};base64,{{ content.input_audio.data }}"
+                            "{% endif %}"
+                        "{% elif content.type == 'text' %}"
                             "{{ content.text }}"
                         "{% endif %}"
                     "{% endfor %}"