[FEATURE] Access to raw response from LLM

[tpaulshippy]

Scope check

• This is core LLM communication (not application logic)
• This benefits most users (not just my use case)
• This can't be solved in application code with current RubyLLM
• I read the Contributing Guide

Due diligence

• I searched existing issues
• I checked the documentation

What problem does this solve?

Reading docs like:

• https://platform.openai.com/docs/guides/rate-limits
• https://docs.anthropic.com/en/api/rate-limits

I'm noticing that these providers respond with quite a bit of rich information about rate limits in the headers of the response.

I also know from working on #234 that there is extra information in the body of the response.

I could also see value in having access to the raw request for analysis which is available on the faraday response object.

Proposed solution

Make the faraday response object available on the message that is returned.

Why this belongs in RubyLLM

Similar to #265 which enabled custom inputs, this would help support custom outputs from the APIs.

[compumike]

I'd personally like this feature because OpenRouter has a Usage Accounting flag that allows you to pass .with_params(usage: {include: true}) and then the response will contain an object:

{
 ...,
 "usage"=>
  {"prompt_tokens"=>16,
   "completion_tokens"=>456,
   "total_tokens"=>472,
   "cost"=>0.0002768,
   "is_byok"=>false,
   "prompt_tokens_details"=>{"cached_tokens"=>0},
   "cost_details"=>{"upstream_inference_cost"=>nil},
   "completion_tokens_details"=>{"reasoning_tokens"=>420}}}
}

and I'd like to track and assign these costs to the corresponding ActiveRecord instance 😄