We would like to allow our clients to send their logs over TCP for
more reliable delivery, and so they will be able to detect if the
central log server is not running.  By default, the imtcp input
module only allows 200 concurrent connections, though, which is a
bit small.  Increase that to 4096, which should be enough for a
long time for NSC (at least as long as worker nodes in our compute
clusters don't send logs directly, but relay via a cluster head
node).

We also need to increase the maximum number of open files (i.e.
the 'ulimit -n' limit) for this to work.