Currently the opus decoder works on the ogg page level, i.e. during
each scheduler iteration it tries to extract one ogg page from the
stream and decodes all opus packages therein. Since ogg pages can be
quite large (20K), decoding can take a significant amount of time,
especially on slow machines. If the decoding takes longer than the
buffer time of the configured writer, audible buffer underruns result.
This changes the opus decoder to decode only a single packet during
each iteration. With this patch applied, latency improves greatly,
and underruns occur on neither an old 150MHz AMD K6 nor on the
raspberry pi.