gcd and lcm on arrays?

[JeffBezanson]

I just noticed that gcd and lcm have methods that reduce them over arrays. That seems like a vestige of implicit vectorization. It does allow gcd to early-out, but I'm not sure it's worth it. Do we want to keep these?

[Keno]

One reason that would argue for keeping it is that I believe it's possible to compute the gcd of an array of numbers faster than reduce(gcd, A), so it might be nice to have a special method for it. Though of course overloading reduce(::typeof(gcd), a) is also possible.

[StefanKarpinski]

Even if that's the case, it seems clearer for the main body to be a proper method of gcd and then defined reduce(::typeof(gcd), a) = gcd(a) to catch cases where someone writes generic code.