Review use of child processes v k8s jobs

I suspect other components could do with a review as well. But particularly for eclcserver I can imagine wanting one eclccserver used for quick jobs - e.g., those that take < 30s, and another for longer running jobs. The first would use child processes, the second would use jobs.

I'm not sure this fits very well into our current scheme though. The eclccservers can't really make the decision - they don't have the information. The queues are based on the thor instances. The closest you could get at the moment is to define two thor clusters thor200 and thor200quick. The second would use a a different eclccserver. That may make sense - you may also want to set the linger option on thor, but I'm not convinced.

The configuration is also currently a bit clunky and could do with discussion and revisiting.