[Feature request] Update to the new model `gpt-realtime`

[stc1988]

New speech-to-speech model gpt-realtime is now generally available.

My understanding is that this includes the following advantages:

• Higher performance and lower cost of use
• two new voices, Cedar and Marin
• Remote MCP server support
• Image Input

I tried using this new model, but there seem to be changes in the message structure, which will require some work.

[stc1988]

As a result of trial and error, I was able to have conversations with the conversational AI app with new voices. I am currently checking the tool function.

[stc1988]

Since we hadn’t been able to continue the conversation after sending the functionResult even before this change, I applied a fix and confirmed it’s working.
For now, I’ll create a PR with the fixes up to this point.