Possible bug in `poisson_weak`

[gaurav-arya]

In the poisson_weak implementation:

adev/src/Numeric/ADEV/Diff.hs

Lines 126 to 133 in 57452f3

	poisson_weak drate = ContT $ \dloss -> do
	let (rate, rate') = split (exp (ln drate))
	x_neg <- poisson rate
	let x_pos = x_neg + 1
	y_neg <- dloss x_neg
	y_pos <- dloss x_pos
	let grad = primal y_pos - primal y_neg
	return (bundle (primal y_neg) (grad * rate'))

it looks like the derivative contribution from the continuation might be missing from the returned dual number. i.e. something like grad * rate' + tangent y_neg rather than just grad * rate'.