taskworker: Use long-lived processpool instead of forked process

[markstory]

• Improve the throughput of the prototypes by using long lived process pools instead of making a new process for each task.
• Improve the pull model by having status updates return the next task. This halves the amount of network requests that pull model needs to make.
• Improve the output of the test harness to make throughput more obvious and format numbers so they are easier to read.