Spliced TCP sockets become faster when the output part is running

Spliced TCP sockets become faster when the output part is running
as its own task thread.  This is inspired by userland copy where a
process also has to go through the scheduler.  This gives the socket
buffer a chance to be filled up and tcp_output() is called less
often and with bigger chunks.
When two kernel tasks share all the workload, the current scheduler
implementation will hang userland processes on single cpu machines.
As a workaround put a yield() into the splicing thread after each
task execution.  This reduces the number of calls of tcp_output()
even more.
OK tedu@ mpi@