Streaming XML decoder can reduce heap memory up to 30%

[jhendrixMSFT]

Current pipeline and codegen will read the response body in the pipeline so that if reading the body fails the operation will be retried. The cached body is then passed to xml.Unmarshal.

It was noticed that skipping body download in the pipeline and swapping out runtime.UnmarshalAsXML for xml.NewDecoder(resp.Body).Decode(&result.ListBlobsFlatSegmentResponse) reduced heap memory consumption by ~30%. Unclear if the same gains would be observed for JSON payloads but should be measured.

I believe we still want the decoding to happen in the pipeline so that if reading the body fails we can retry the request. This would likely require some new APIs in azcore and slight refactoring of codegen such that the *HandleResponse methods perform the decoding as part of the pipeline.