Speed up counting in statistics.fmean()

[rhettinger]

Implement a faster counting strategy that avoids tuple creation, indexing, and destruction.

Baseline timing:

% ./python.exe -m timeit -s 'from statistics import fmean' 'fmean(iter(range(10**3)))'
20000 loops, best of 5: 18.7 usec per loop

After the PR:

% ./python.exe -m timeit -s 'from statistics import fmean' 'fmean(iter(range(10**3)))'
20000 loops, best of 5: 11.4 usec per loop

[eendebakpt]

Nice trick!

I like the current PR, but just for fun: the counter still produces an int on every iteration. Much faster is:

            counter = repeat(True, 2**60)       
            total = fsum(compress(data, counter))            
            n = 2**60 - counter.__reduce_ex__(4)[1][1]           

(older python version, the __reduce_ex__ support has been removed and I cannot find any other object with an internal counter)