TODO

Things to do to cfile:

improvements:
* cfsize returns 0 on a file being written?  Or keep a track of how
  many bytes we've written and return that.
* cfile can also provide cfgetc
* providing cfungetc seems pointless though - who uses ungetc?
* Can we read the size of a BZIP2 file by finding and counting each block in 
  turn?  Presumably each block knows how much compressed
  data it contains so it can find the next block, and how much uncompressed 
  data it contains because it has to do the Burrows-Wheeler Transform on a fixed size.
  Skipping through blocks like that would be quicker than reading the whole
  file...  Maybe talk to Julian Seward about this?