Add documentation and example for feedback reward leveraging custom metrics

[axsaucedo]

Based on the core workstream focused on basic statistical metrics, the first task will be to showcase the capabilities of what can be done with today's model capabilities. This task will consist of the following:

• Adding an example that shows how some of this can be achieved today with custom metrics - namely tracking accuracy using reward as 0=incorrect 1=correct (this is dependent on metrics changes above) + general documentation on what is possible today