chagne frontend mode, output a json dict, output to stdout by default, output duration with stats, not only once

So look for those, and concatenate the lines, so we don't strip the control
information from the first one only. Not sure if that's per spec, but I've
got one sample with those, and it's from the VLC test streams, so it must
be more than just an oddball.

 data which wasn't initialized when the init code was compiled out.

* use av_audio_resample_init
* new option to output information about files as json array

track audio and video eos separately so they are closer to each other if -s / -e is used to cut out segments

  if seeking between keyframes

  Only text (raw UTF-8 and SSA) is supported.
  SSA tags are not properly removed yet.
* update libkate to 0.3.1