when reading frames, need to read a bit extra in order to have the
headers of the frame following the last frame of the segment (in order
to calcuation the duration of the last frame)
in addition to reading a few bytes extra, need to ignore overflow when
parsing the cluster/simple-block elements of the next-segment frame,
since usually this frame is not read in full.

- a block group is a key frame if it has no block references (unlike
simple groups which use flags & 0x80)
- also improved debug logs - added per-lace debug log, including the
  absolute file offset, size, key frame etc

can significantly reduce the read overhead - on some sample content that
was tested, the read overhead (=total_frame_size_read - total_output_size)
was reduced by 44%.