if the entire soundfile fits into the fifo, then [readsf~] will start
to loop the file, since the 'drained' test in d_soundfile.c:1937 does
not trigger.
the fix is to properly set the fifohead (in the read thread) to the new
position, even if the read() returned '0' (indicating EOF)

commit 057cb3ee6cc58597a1b74a00a7880c2cfd0a466a