[EPIC] Abstractions for monitoring outputs

[shreyashankar]

We want to monitor ML pipelines to see when things might be going wrong. We need a way to encode "SLAs" into mltrace, then monitor outputs and metrics to make sure SLAs are met.

To Dos:

Logging

• Add 2 new tables (outputs, feedback) that stores ts, key, val, identifiers/labels, indexed on ts, key
• Add functions to write and read from those tables

Querying

• Add function that joins the 2 tables
• Add functionality to define metrics (i.e. accuracy, precision, recall, f1) to run on the above joined table #283

Interface

• Metric class with log_pred, log_true and compute_metric functions
• "Monitor" view that accepts the group name(s), window size, and metric name
• chartjs graph to show the above