Percentile is innacurate with a small data set

[scott-david-walker]

When calling Query using a small data set the actual underlying call to calculate the percentile is never used unless the stream has been flushed. It appears just to return the value in a certain index.

	q := quantile.NewTargeted(0.5, 0.9, 0.95, 0.99)
	q.Insert(float64(5340000000))
	q.Insert(float64(5750000000))
	q.Insert(float64(5930000000))
	q.Insert(float64(6160000000))
	q.Insert(float64(6560000000))
	q.Insert(float64(1156000000))
	fmt.Print(q.Query(0.50))  // 5750000000
	fmt.Print(q.Query(0.90)) // 6160000000
	fmt.Print(q.Query(0.95)) // 6160000000
	fmt.Print(q.Query(0.99)) // 6160000000

[bmizerany]

It's been a while since I've been active with this project and code, but IIRC, this is to be expected. The sketch is for larger datasets. Datasets that are bounded and that fit in memory are probably best off using simpler, more accurate ways of computing quantiles.