Collect basic performance metrics

[yannham]

As #1622 shows, it's not easy to profile the Nickel interpreter: most time is spend inserting and fetching from the environment, evaluating primitive operations, but this is what is expected from a Nickel program.

In order to better diagnose performance issues, we should add the required infrastructure to collect basic Nickel-specific metrics in the interpreter - insertion and fetch from the environment, average size of environments, environment cloning, thunks created, thunks forced, generated variables created, etc. - and a feature flag to enable collection and report.

[yannham]

Closed by #1633 #1635. Opened for tracking purpose and future reference.