konq_run / krun should determine the mimetype by actually
getting the contents of the URL. It should then put the slave
on hold and tell the job-scheduler which request the
slave is currently handling. (Status: implemented in konq_run)

Now krun/konq_run should determine which client should process the 
result of the request.

* When the client belongs to the same process, no action needs to be
taken. When a new job is created for the request which is on hold the
existing slave will be re-used and the request resumed.
(Status: implemented)

* When the client is an external process, the on-hold-slave should be 
removed from the job-scheduler and should connect itself with
tdelauncher. This is hard because it must ensure that the external 
program does not request the slave before it has been transfered to
tdelauncher. (Status: to be done)

* When a slave is on hold but not used for a certain period of time, 
or, when another slave is put on hold, the slave should be killed.
(Status: almost done)

=====

The slave must emit "mimetype" during a GET before the first data is send.

It may wait with sending "mimetype" until it has enough data to 
determine the mimetype, but it should not pass any data along before it has
send the mimetype.

Currently only http _always_ sends a mimetype.