TTSVoice type is missing currently supported OpenAI voices #4534

Describe the bug

TTSModelSettings.voice is typed as TTSVoice, but the current TTSVoice literal is missing OpenAI speech voices that are supported by the underlying OpenAI Python client: ballad, verse, marin, and cedar.

As a result, valid OpenAI TTS configurations using those built-in voices are rejected by static type checkers even though OpenAITTSModel forwards the value to audio.speech.create() and the OpenAI API supports it.

Current type

TTSVoice = Literal[
    "alloy", "ash", "coral", "echo", "fable", "onyx", "nova", "sage", "shimmer"
]

Expected behavior

The Agents SDK's built-in TTS voice type should include the currently supported built-in OpenAI voices, including ballad, verse, marin, and cedar.

Scope

This only requires updating the public voice literal; runtime request behavior is already capable of forwarding these string values.