generic worker should handle OOM better

[bhearsum]

At the moment, if a task runs something that consumes all of the system RAM, generic-worker tends to get OOM-killed, resulting in the task failing with CLAIM_EXPIRED. This is a bit unfortunate, because this state can happen for a variety of reasons.

Ideally, we'd avoid generic-worker getting OOM-killed entirely (@petemoore - I believe you have ideas about this).

A smaller measure, if possible, might be to detect the OOM kill, and either log a clear message to the log or resolve the task with a particular state.