check for iszero(partials(x)) in pow

[chriselrod]

No description provided.

[chriselrod]

As suggested by @andreasnoack, I think it's also worth considering StatsFuns.xlogy in place of x * log(y) within ForwardDiff.

That'd be a bigger change, so it's better suited for a separate PR. I just mention it here on the theme of better handling of 0.

Note that y ^ x in Base is implemented via an extended precision exp(x * log(y)), minimal code should be needed for an accurate x * log(y).

[devmotion]

As suggested by @andreasnoack, I think it's also worth considering StatsFuns.xlogy in place of x * log(y) within ForwardDiff.

As you might know, StatsFuns just reexports LogExpFunctions.xlogy, and since ForwardDiff already depends on LogExpFunctions this would not even require any additional dependencies.