Minor improvement to statistics.pdf()

[rhettinger]

The current code squares sigma and then takes the square root of the result. The unnecessary round-trip increases overflow/underflow risks, costs a little time, and loses a tiny bit of numerical accuracy.

Avoid the round trip by normalizing the data first.

% python3.15 -m timeit -s 'from statistics import NormalDist as ND' -s 'iq=ND(100,15)' 'iq.pdf(121.1)'
2000000 loops, best of 5: 121 nsec per loop

% python3.15 -m timeit -s 'from statistics import NormalDist as ND' -s 'iq=ND(100,15)' 'iq.pdf2(121.1)'
5000000 loops, best of 5: 99.5 nsec per loop

This also makes the error message slightly more accurate. Formerly, a subnormal sigma could underflow the variance to zero and the message would report that sigma itself was zero.