* Fix a nasty problem with output buffering not have enough (or large enough)

  buffers. In fact, the code was pretty much shot.
* Try to fool WMP into thinking that we are a microsoft server.
* When we establish a stream to a user, copy the codec information from
  that saved as part of the stream. This gives us the real frame_size and
  other important parameters.
* ASF needs to know about key frames, so add some logic to copy this information
  around.
* When we get the data from ffmpeg as part of a feed, make sure that we save
  the actual codec parameters.
* Allow configuration of AudioCodec and VideoCodec
* Make sure that we delete the feed file before starting. This is not ideal
  but it makes things work a whole lot better!

Originally committed as revision 454 to svn://svn.ffmpeg.org/ffmpeg/trunk