AppServerMetrics reporting-thread fix

[wjordan]

This PR resolves an issue with the AppServerMetrics metrics-reporting thread being shutdown by Puma, by manually spawning the reporting task in a before_block instead of implicitly when the middleware is instantiated.

The issue stems from a slight apparent difference between Puma and Unicorn in the daemonization process. While Unicorn preserves threads created by the master process, Puma shuts down any threads spawned during the app-preload process unless they are created after daemonize! occurs within the before_fork hook.

In order to access the middleware instance within the hook, I added an instance class-accessor to the middleware class. This doesn't seem ideal, but I couldn't figure any better way to access the middleware instance (since the instance is created dynamically from the Rack builder when the middleware stack is built on startup, there's no place in code we can assign the individual instance for later access). This shouldn't be a problem in practice, since there should only be a single instance of this middleware-class created per process.

[wjordan]

Confirmed on an adhoc instance that this PR results in correct metrics being reported to CloudWatch.