Come up with improvement to numbers/mixtureModel.go stdev threshold

[craiglowe]

		case mm.Stdev[i] < 1e-04: // most likely caused by overfit of gaussian to a single data point

Should this be a relative instead of absolute threshold?