expm1 should having float inputs

[honno]

Was expm1() accepting int dtypes an oversight? If so, here's a fix!

If not, I'm guessing the intention was to "allow" int dtypes here, but leave it as implementation-defined behaviour. In that case we'll need to add a note ala divide().

[asmeurer]

Presumably this was just a typo. expm1 should be no different from exp. And it looks like the test suite and numpy implementation are already correct here.

[asmeurer]

All the floating-point functions can accept integer inputs if the library wants to. I don't think we need such a note. I think divide only gets a note because it's not obvious whether the spec requires integer division on integer inputs or not (compared to something like exp where it would be pointless to support integer inputs unless you did a float promotion).

[kgryte]

This appears to have been an oversight. As such, will merge...