Fix misleading error message while streaming

[Wauplin]

In innerRequest/innerStreamingRequest, if an error occur we raise a InferenceClientProviderApiError with a HTTP Request object in it. To avoid bloating the logs, a requestArgsToJson helper is applied to clean-up the body. However we do not pass the body to its but the requestArgs instead i.e. before provider-specific logic has been applied. This is can be really misleading since the body shown in the error is not necessarily the one sent in the request, making it hard to investigate.

I also took the opportunity to make sure accessToken is never printed in the logs (it is currently the case as part of requestArgs).

(also added a console.log for auto-provider)

[Wauplin]

Example of misleading stracktrace:

    at async postCreateResponse (/Users/vb/Documents/hf-projects/responses.js/src/routes/responses.ts:185:21) {
  httpRequest: {
    url: 'https://router.huggingface.co/novita/v3/openai/chat/completions',
    method: 'POST',
    headers: {
      Authorization: 'Bearer [redacted]',
      'Content-Type': 'application/json',
      'User-Agent': '@huggingface/inference/4.3.0 Node.js/23'
    },
    body: {
      accessToken: '', # <=========== token leaked here + shouldn't be in body
      model: 'meta-llama/Llama-3.3-70B-Instruct',
      messages: [Array],
      stream: true,
      temperature: 1,
      tools: [Array],
      top_p: 1,
      data: null  # <================= shouldn't be in body
    }
  },
  httpResponse: {
    requestId: 'Root=1-686651cf-4f0ed45361744c1734cb3c1f',
    status: 404,
    body: { message: 'API error', type: 'API_ERROR' }
  }
}

[hanouticelina]

LGTM! Thank you

[SBrandeis]

good catch !!

[Wauplin]

thanks for the quick review :) I'll make a patch release