Expose raw provider response (including custom fields) in Spring AI client responses

Summary

When using Spring AI with custom providers (or providers that include additional
fields such as usage cost, billing, model metadata, etc.), the framework does
not currently expose the raw provider response or a structured way to access
custom fields that are not mapped into Spring AI’s default response models.

This limits the ability to read provider-specific metadata (e.g., cost fields)
that exist in the underlying API response but are not mapped by Spring AI.

Problem

Many LLM providers return additional fields in their JSON response.
Examples:

• token-level cost breakdown
• provider usage metadata
• custom domain-specific values
• provider-specific debugging fields

Spring AI currently:

• parses responses into internal domain classes (ChatResponse, Generation,
  Metadata, etc.)
• but drops unrecognized fields
• and does not expose the raw JSON returned by the provider

Because of this, there is currently no way to access provider-specific
metadata, even via Metadata, Advisors, or Interceptors.

What I Tried

• Using Advisors (CallAdvisor, StreamAdvisorChain, etc.) — they provide
  access only to Spring AI domain models, not raw JSON.
• Custom Observation handlers — still no access to unmapped fields.
• Looking for Interceptors in the OpenAI client — not available.
• Checking if Metadata preserves unknown fields — it does not.

Requested Feature

I’d like Spring AI to provide one or both of the following:

1. Access to the raw provider response

For example:

ChatResponse response = client.call(prompt);
String rawJson = response.rawResponse();

Start by reviewing ChatResponse, Generation, Metadata, Advisors, and Interceptors to understand where provider responses are mapped and where unmapped fields are lost. Decide which requested raw-response or custom-metadata API is appropriate, then verify that provider-specific fields remain accessible through the resulting client response.