On truncated jpeg, decoder can suspend waiting for additional bytes in
buffer. For some input files, decoder suspends on jpeg_start_decompress
stage. If at this point file reader reaches EOF, py code never gets back
to jpeg decoder and we end up with no bytes to result image. This leaves
us with some amount of potentially useful bytes undecoded and thrown
away.
Libjpeg docs suggest that in such situation, more appropriate would be
to add EOI marker to the end of buffer, which will allows decoder
to finish. https://github.com/libjpeg-turbo/libjpeg-turbo/blob/0dd9a2c1fd6c/libjpeg.txt#L1803-L1809
Docs also mention that adding EOI markers is what non-suspending code
does anyway.
