Add Human Evaluation from inside Evalite

[mattpocock]

The idea is:

• You would specify a special humanScorer() in the scores array passed to evalite()
• In the UI, you could mark the result with a tick or cross
• This would be saved to a local JSON file, keyed under the eval name, input, expected and output. (this means that if the input, expected or output changes, the human score would be invalidated)
• The UI would then retrieve this JSON information where needed to display the scores.

[mattpocock]

More thoughts here:

• Having a step-by-step review process would be really useful for lots of stuff I'm building (content production)
• You click 'review mode' and it puts you into a custom UI where you can use keyboard shortcuts to navigate and quickly blast through reviews
• Human reviews are cached using Storage.