Lazily initialize RAPL wraparound monitor processes

[jaywonchung]

Currently, whenever get_cpus is called, it will instantiate a RAPLFile for every CPU/DRAM domain, which will in turn spawn a wraparound monitor process. However, the user may not intend to use RAPL anyway, which makes the processes poll RAPL for nothing.

One good way is to start those processes lazily, meaning that starting the wraparound monitor processes is deferred as much as possible.
That is, only when RAPL metrics are actually needed (e.g., ZeusMonitor that tracks a non-empty cpu_indices calls begin_window), the processes will be spawned only then.

@wbjin WDYT?

[wbjin]

I agree, we could wait until the very first read() on a RAPLFile to start the monitoring process. I think we could do something like this
Initialize self.wraparound_tracker = None.
In read() initialize RaplWraparoundTracker if self.wraparound_tracker == None