Add *new_only* parameter for process_iter()

[giampaolo]

When true, process_iter() will yield new process only (meaning new PIDs appeared since last call):

>>> import psutil
>>> for proc in psutil.process_iter(attrs=['pid', 'name'], new_only=True):
...     print(proc.info)
...

This comes from a real-world use case I bumped into: I spawn a subprocess but don't know its PID, so I look for it by checking the command line of all processes. Instead of iterating over all processes, I want to iterate over new PIDs only (faster).

[giampaolo]

On a second thought, I realized that process_iter() uses a global variable, so it's not thread safe.
Despite this is acceptable when using it in the standard way, this case is different: if the are 2 threads using process_iter(), the first one calling it (+ consuming the iterator) will "steal" the processes of the second thread. That aside, it's unclear how to solve this problem in terms of API, so I reverted this PR in c9fc4fd. Problem solved.

We have another one though: psutil.cpu_percent(). Since it also uses a global var, multiple threads will "reset" the internal timings of each other and invalidate result. Not sure why I realized this only now but the doc should be more clear about this.

Anyway, I'll discuss this in another ticket.