jackknifeVarianceUnb fails for input of length 2

[patrickdoc]

While investigating Criterion issue #162, I've uncovered an problem with the jackknife calculations.

https://github.com/bos/statistics/blob/98a5e888ab411b3f48d31bb677b4c8f1db880217/Statistics/Resampling.hs#L223-L245

When the input is of length 2, q is 1. For jackknifeVarianceUnb, c will also be 1. This means that at the very least, go will divide everything by 0.

Additionally, if my algebra skills haven't failed me, al + ar - (b * b) will always be 0, so go is just (0/1)/0 = NaN for both elements of the sample.

[Shimuuar]

Thank you! jackknife works by throwing out every other sample and if sample have two elements jackknife will try to estimate variance for two one-element sample and will fail.

I'll add call to error for 2-element sample and variance/standard deviation. But it's part of more general problem (#100)

[Shimuuar]

fixed in d842eb7