*** fix for read()

Reading 1-char per read() is the less intrusive way to fix the behaviour I observed.
Not a nice fix, but the client requests shouldn't steal so much bandwidth
*** fix for find_rtp_session_with_url.
Note that mplayer send one PLAY request per stream, I don't know if this should be handled
more gracefully
patch by (Giancarlo Formicuccia <ilsensine at inwind dot it>)

Originally committed as revision 1998 to svn://svn.ffmpeg.org/ffmpeg/trunk