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