Retry-After headers when 'CreateChatCompletionStream' fails due rate limit hits

[kanutron]

I believe this could be a duplicate of #452, but the discussion there diverged from the actual point.

Problem

When using the CreateChatCompletionStream, and hitting rate limits errors (Code: 429), the returned error nor the stream doesn't contain the http headers. I need to retrieve 'Retry-After' headers to pause the job for the specified time.

Those headers are standard and provide the amount of time (seconds/time-stamp on 'Retry-After', milliseconds on 'Retry-After-Ms').

If a 429 is obtained, what I do with CreateChatCompletion is to parse those headers and pause to go routine for the specified amount of time.

	resp, err := client.client.CreateChatCompletion(ctx, *openaiReq)
	if err != nil {
		... check status code ...
		parseAndPause(resp.Header()) // even with error, we still get `resp`
	}

But for CreateChatCompletionStream the above code could not work as:

	stream, err := client.client.CreateChatCompletionStream(ctx, *openaiReq)
	if err != nil {
		... check status code ...
		parseAndPause(stream.Header()) // stream.header could be nil, thus this panics
	}

Proposed solution

I believe here:

https://github.com/sashabaranov/go-openai/blob/7c145ebb4be68610bc3bb5377b754944307d44fd/client.go#L166C1-L172C3

	resp, err := client.config.HTTPClient.Do(req) //nolint:bodyclose // body is closed in stream.Close()
	if err != nil {
		return new(streamReader[T]), err
	}
	if isFailureStatusCode(resp) {
		return new(streamReader[T]), client.handleErrorResp(resp)
	}

Those returns should add the response into the streamReader struct.