netxlite: consider bypassing net.Resolver completely

[bassosimone]

This issue is about bypassing net.Resolver in internal/netxlite when compiled with CGO_ENABLED=0 (which is what one would get when cross compiling the code). The ooni/probe-cli#764 pull request is changing netxlite to bypass net.Resolver with CGO_ENABLED=1 in favor of calling getaddrinfo directly to fix #2029. This means that we will always have a working resolver when compiled with CGO_ENABLED=1. What's more, we'll have the luxury of using the system resolver, which reportedly is the only reliable resolver in some very filtered networks (e.g., during partial shutdowns). OTOH, with CGO_ENABLED=0, we are still using net.Resolver in its netgo flavor, which means reading /etc/resolv.conf. Because there is no /etc/resolv.conf on Android, miniooni cross compiled for Android does not work in this scenario (this is already the case regardless of our changes to improve getaddrinfo). My proposal here is to fallback to an UDP resolver in this case rather than relying on using the one written by the Go time and included in the stdlib. Such a resolver would probably attempt to use /etc/resolv.conf along with a set of hardcoded resolvers (e.g., 8.8.8.8:53/udp) trying to provide CGO_ENABLED=0 users with a workable resolver. I'm flagging this issue as a user request because we received reports of miniooni not working on Android using Termux in Slack.