Fixes:
1. Use ogg_int64_t to calculate and store presentation times of
   frames, rather than transitioning through double just to cast back
   to ogg_int64_t. Prevents floating point rounding errors from
   making frames sometimes off by 1ms.
2. Calculate the duration of vorbis packets using the vorbis
   block/window data. This means that we can accurately know what
   presentation time we'll get if we decode a vorbis packet. Using
   this and the knowledge that vorbis packets require their preceding
   packet to decode, we can now accurately calculate the presentation
   time we'd get if we began decoding from a vorbis page (rather than
   just taking the previous page's end-time, which doesn't take into
   account the fact that we can't decode the first vorbis packet).