Support multimodality in chat completion output

[ThomasVitale]

OpenAI has recently introduced audio multimodality support, both for input and output.

The input audio modality support is introduced in #1560 all the way up to the Spring AI abstractions.

The output audio modality is only supported at the lower level (OpenAIApi). Its usage is demonstrated in this integration test:

spring-ai/models/spring-ai-openai/src/test/java/org/springframework/ai/openai/api/OpenAiApiIT.java

Lines 98 to 118 in bdb66e5

	@Test
	void outputAudio() {
	ChatCompletionMessage chatCompletionMessage = new ChatCompletionMessage(
	"What is the magic spell to make objects fly?", Role.USER);
	ChatCompletionRequest.AudioParameters audioParameters = new ChatCompletionRequest.AudioParameters(
	ChatCompletionRequest.AudioParameters.Voice.NOVA,
	ChatCompletionRequest.AudioParameters.AudioResponseFormat.MP3);
	ChatCompletionRequest chatCompletionRequest = new ChatCompletionRequest(List.of(chatCompletionMessage),
	OpenAiApi.ChatModel.GPT_4_O_AUDIO_PREVIEW.getValue(), audioParameters);
	ResponseEntity<ChatCompletion> response = openAiApi.chatCompletionEntity(chatCompletionRequest);
	assertThat(response).isNotNull();
	assertThat(response.getBody()).isNotNull();
	assertThat(response.getBody().usage().promptTokenDetails().audioTokens()).isEqualTo(0);
	assertThat(response.getBody().usage().completionTokenDetails().audioTokens()).isGreaterThan(0);
	assertThat(response.getBody().choices().get(0).message().audioOutput().data()).isNotNull();
	assertThat(response.getBody().choices().get(0).message().audioOutput().transcript())
	.containsIgnoringCase("leviosa");
	}

It would be nice to start identifying what type of abstractions are needed in the ChatResponse API to include audio response data.