fix: Recover periodic metric readers after forking

[chrisholmes]

This PR provides a fix to PeriodicMetricReader so that it continues working after a Ruby process is forked. This problem currently surfaces itself on applications that use forking webservers.

When forking a Ruby process only the thread creating the child process is coped across from the parent process, which means that background threads need to be created for monitoring tools to continue working. This means that, before this fix, PeriodicMetricReader would not work on the child process as the background thread is not copied.

The fix is implemented by overriding the Process._fork method provided in Ruby versions 3.1+. Overriding this method allows the authors of monitoring tools to introduce callbacks on fork events.

An alternative approach, required prior to 3.1, would require clients to call hooks directly from their forking webserver's configuration such as use on_worker_boot in puma.rb. Overriding Process._fork removes the need for this boilerplate.

When resetting the PeriodicMetricReader, the after_fork hook will first read and discard metrics that would have been published by the parent process. This is in order to avoid duplicate publishing of metrics that will be handled by the parent process.

The parent process' PeriodicMetricReader is left running after forking as it is assumed that clients will want to publish metrics from it still.