Choose resolvers at random instead of always taking the first resolver

[jcgruenhage]

Most people will be using public_ip::addr (or one of the address family specific subvariants), which will always take the first enabled resolver, falling back to other enabled resolvers until it finds one that can complete the request.

To me, it would make more sense to shuffle the list of providers, so that the load is balanced among the different providers here.

I'd be willing to implement this, but I'd like to wait for feedback here if this is wanted and whether you have any strong opinions on how this should be implemented.