Accept both Cvoid and Ptr{Void} return type in ccall(:memcpy)

[Keno]

POSIX specifies that memcpy returns its first argument, which is
a bit of a useless feature (e.g. the llvm intrinsic just returns
void). Nevertheless, since we're intercepting the ccall here, we
should allow it. For convenience, still allow the Cvoid return though.

Fixes #31073

[vtjnash]

I'm not sure how much we want to commit to guaranteeing that this precise input to ccall is allowed to be incorrect with respect to the C function that is being called. If we keep the code to permit return of void, then we may need to also document that it has this one special case. (although likely on most platforms, return of a pointer vs. return void probably doesn't affect the caller ABI, so it'll likely work even though wrong)

[Keno]

Ok, I don't have any particularly strong preference either way. However, if we fixup the call site and switch this around, we'll probably have to do a different backport, since otherwise we'd be causing performance problems in the backports if anybody relied on the old signature. Up to you.

[KristofferC]

Bump

[JeffBezanson]

I think we should merge this. Might as well tolerate as much stuff as possible :)