12614: Detect and report broken node errors.
1. crunch-run has an errorBlacklist of messages that mean "this Node probably
cannot run any container successfully"
2. Upon seeing a blacklisted error, the broken node hook script is run
3. The hook script is responsible for marking the node as "broken, don't
schedule work on it, such as executing scontrol reconfigure Node=XXX State=DOWN
4. After the hook script terminates, if possible the container is reset from Locked
back to Queued state.
5. Then crunch-run can terminate.
Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <pamstutz@veritasgenetics.com>