I would assume that JPEG like most lossy compression algorithms is specifically defined in terms of the correct uncompressed output for a given compressed input. This means a "deblocking JPEG decoder" is effectively actually just a bad decoder that gives the wrong output (albeit for some outputs maybe you happen to prefer this).

The definitions are this way around because it's the only way to have an opportunity to improve encoding quality over time, if you start by defining how encoding must be done then you lock in bad ideas. Lame for example is far, far better at encoding MPEG audio layer III than the demo encoders shipped when the standard was written, and this is only possible because the standard says how to turn MP3 back into PCM data, not how to encode PCM data as MP3.