feat: support OpenAI response continuation

[arnabnandy7]

Summary

Add OpenAI Responses API conversation continuation support using previous_response_id.

Changes

• Forward previous_response_id from OpenAiChatOptions.extraBody.
• Add OpenAiResponsesOptions.PREVIOUS_RESPONSE_ID as the supported option key.
• Expose the returned OpenAI response ID through ChatResponseMetadata.id.
• Document how to persist and reuse response IDs.
• Clarify that callers should send only new input when continuing a stored conversation.
• Add unit coverage for request and response ID mapping.

Example

val firstResponse = model.call(Prompt("Tell me a short story"))

val nextOptions = OpenAiChatOptions.builder()
    .model("gpt-5-pro")
    .extraBody(
        mapOf(
            OpenAiResponsesOptions.PREVIOUS_RESPONSE_ID to firstResponse.metadata.id
        )
    )
    .build()

val nextResponse = model.call(
    Prompt(
        listOf(UserMessage("Give it a happier ending")),
        nextOptions,
    )
)

Scope

This change supports server-side conversation continuation for models routed through the Responses API. It does not enable server-side tool execution. Embabel continues to own the tool loop.

Testing

./mvnw -pl embabel-agent-autoconfigure/models/embabel-agent-openai-autoconfigure `
  -Dtest=OpenAiResponsesChatModelTest test

Closes #1610.