Worker pool stats broken

When a worker is checked out from a pool (i.e. through a call to riak_core_worker_pool:poolboy_checkout/3 the timestamp of the checkout is updated on the list of checkouts. A queue_time stat is updated by comparing the time the job was added to the queue with now.

When a worker is checked back into the pool (i.e. through a call to riak_core_worker_pool:poolboy_checkin/4, a work_time stat is updated by comparing the checkout time for this pid with now.

However, if work is completed, and there is work queued handle_event({checkin, Worker} bypasses the poolboy_checkin/4 and poolboycheckout/3 functions, and simply allocates the next work in the queue directly to the checked in process.

This means when there is a large queue, and workers are continually recycled fresh work - the stats no longer make sense. For example, only when the queue is emptied will the work_time be recorded, and this will not be the time for the individual piece of work but the time since the worker first became busy.

Start by tracing riak_core_worker_pool:poolboy_checkout/3, poolboy_checkin/4, poolboycheckout/3, and handle_event({checkin, Worker}). Reproduce a continuously queued workload and inspect how checkout, queue_time, and work_time are updated. Done means the stats represent each individual queued job and worker handoff, including when work is recycled directly.