Detection of <Error><Code> in the body: Even if error_code == AWS_ERROR_SUCCESS (HTTP 200), we parse the XML body to look for an <Error><Code>. If found, we map the S3 error to a CRT error code (or AWS_ERROR_S3_NON_RECOVERABLE_ASYNC_ERROR if unknown). This is an extra safety net in addition to the generic mechanism in s3_meta_request.c, which already performs this check beforehand but might miss it in some timing cases (body arriving after send_request_finish).

Verification of the ETag presence: On a real success, the body must contain a <CompleteMultipartUploadResult> with an <ETag>. If we get a 200 but no ETag (empty body, truncated body, or unexpected content), it must be an issue → we force AWS_ERROR_S3_NON_RECOVERABLE_ASYNC_ERROR.

Conditional invocation of the headers_callback: Before the patch, the user callback (headers_callback) was always invoked on a 200, even if the body contained an error. Now it is only called if error_code is still AWS_ERROR_SUCCESS after the two checks above. This prevents notifying the caller of a "success" that is not actually one.