Expand evaluation doc

[alextrott16]

Uses evaluation logic pseudocode to provide a better intuition for the logic and the relationship between ComposerModel.validate and ComposerModel.metrics.

Adds some exposition around using the Evaluators class.

Closes CO-861

[alextrott16]

@hanlint Not sure if this addresses all the aspects of CO-861, but AFAICT the validate method should be the same regardless of what style of metric it feeds into (i.e., the output tuple from validate will always be used as the positional arguments for metric.update regardless of whether it represents a single metric or a list-style or dict-style metric collection).