Backport per-request custom header application for openai-sdk

In https://github.com/spring-projects/spring-ai/commit/f36dce3bd22c9e20159fc715d23a2da88b610f1e included in the current 2.0.0-M5, the application of custom headers from prompt chat options was added for openai(-sdk) based chat models.

/*
 * org.springframework.ai.openaisdk.OpenAiSdkChatModel#createRequest (2.0.0-M5)
 */
ChatCompletionCreateParams createRequest(Prompt prompt, boolean stream) {

  // ...
  
  if (requestOptions.getCustomHeaders() != null && !requestOptions.getCustomHeaders().isEmpty()) {
      requestOptions.getCustomHeaders().forEach(builder::putAdditionalHeader);
  }

  // ...

This is a very useful and handy addition, because it allows things like per-request authorization request headers with only a single chat model/client.

final String content = chatClient
    .prompt(Prompt.builder()
        .content("How are you?")
        .chatOptions(OpenAiChatOptions.builder().customHeaders(Map.of(
                HttpHeaders.AUTHORIZATION, "Bearer some-tenant-specific-key"
            ))
            .build())
        .build())
    .call();

Because this is such a useful change I would kindly request to backport it to 1.1.6.

Start by comparing the 1.1.6 openai(-sdk) chat model implementation with org.springframework.ai.openaisdk.OpenAiSdkChatModel#createRequest and commit f36dce3bd22c9e20159fc715d23a2da88b610f1e. Trace how Prompt chat options reach the request builder, then verify that custom headers are applied per request in the 1.1.6 code path.