HowTo: error diagnosis, and recovery

[hjoliver]

New users often don't know where to look when things go wrong.

It might be a while before all the relevant info gets exposed in the GUI, and even then we should document lower level process.

1. check the scheduler log, to see if the scheduler encountered a problem, or if a task did
2. if a task failed in execution, check its job.out and job.err
3. if a task failed job submission, check job-activity.log

And document how recover from common kinds of failure. (Partially dependent on near-future development of cylc set-outputs etc.)

[oliver-sanders]

Good idea, some kind of troubleshooting section.