We now run jobs w/o limits. This is probably bad in a hosted setup but
we can worry about that later should it ever become a problem. As it
stands the implementation is easier with a simple goroutine rather than
a complicated pool of threads.