Add ping.py so we can test when a host is up. ping -c 2 -w 90 will
almost do the right thing, but will exit on a icmp host unreachable.
We need to continue to wait. ping.py just waits for an answer to a
ping.
Parallelize building the compute nodes (but not the head node).
Kill ping process in ping.py.
Copy PIP cache to host before running.
Add syslog artifacts.
Remove known_host weirdness, depend on .ssh/config instead.
Change-Id: I65a12074cf80feb3ae69d15c81ba32a81566b22a
7874c0cec0