Voice parameter in SpeechRequest to be a free form text param instead of restricted enum

Expected Behavior

We should be able to use OpenAI api with other services that are compatible with OpenAI API contract. In this case, voice parameter to be free text so server will interpret that variable however it wants rather than artificially restricting it to an enum.

Current Behavior

As of now, voice parameter is an enum. So, we cant use the API with other services that are OpenAI compatible unless they too support exact same options.

Context

Voice should be a freeform text so OpenAI API can be used with all OpenAI compatible services. For e.g, tools like kokoro needs user to specify the voice thats not provided by OpenAI i.e kokuru has its own mechanism to define voices rather than standard formats that OpenAI supports. One of the value for voice being "af_heart" & others being "af_sky+af_bella" i.e mix of two voices.

By keeping the variable as a string/Voice type that get marshalled into string when being sent to server, new services can use it however they want rather than being artificially narrowed down to just OpenAI provided speech formats (may be these options OpenAI currently supports increase tomorrow).

Because this is not supported in Spring AI, I am using another library that allows me to specify freetext instead.

Start at the SpeechRequest voice parameter and trace how its type is serialized for the OpenAI-compatible API. Check existing SpeechRequest tests, if present, and verify that arbitrary voice strings such as "af_heart" and "af_sky+af_bella" are accepted and sent unchanged.