A slow upload is told as slow, not as our timeout

A request whose body is still arriving when the API's 110-second answer deadline passes, such as a chunk of a large upload over a busy link, is now answered 408 request_timeout with the bytes received so far, where before it was answered 503 api_unavailable, as if the API were at fault. The SDKs and the CLI get 503 body_stalled and send it again by themselves, as they did before. Many uploads at once over one slow link share it: send fewer at once, or send again with the same Idempotency-Key (a chunk at its same offset). See errors.