Add data validation layer to detect "obviously incorrect" data output

[erichte-ibm]

For example, a machine should never report cpu.cores = 0. Perhaps there should be some level of validation either in the collector, or across all collectors to try to detect when a command may have failed, or a command may not be implemented correctly on a platform.

[erichte-ibm]

This should probably be handled by the collectors themselves, and reported as an error. Perhaps we need a new error return type for these cases?