2. Go does not have "a powerful GC", it has a GC built quite specifically for it and the applications it's intended to support with low-latency, middling throughput, and the assumption that allocations are long-lived (because short-lived objects don't escape and thus don't get heap-allocated)
3. Which is the issue of providing generic GC strategies: the type of GC you want and how you want to tune it is very much a factor of the language's semantics and its intended use cases, hence the official JVM shipping with something like half a dozen different GCs, and third-party JVMs having their own with their own tradeoffs
And when one have some (parts, modules, libraries) in your program which are best suited with different GC behaviour, things tend to become complicated. And cross-language boundaries is yet another talk.
2. The assumption that allocations are long-lived can often be made because a compiler can perform escape-analysis (and in fact LLVM could provide utilities for this).
3. Yes, I'm not saying that GoLang's GC (or an adaptation thereof) is the holy grail, just that it would be very useful for many other languages. And work in this area could help bring LLVM's suite of GCs to a competitive, world-class level.
That being said, the Bohem garbage collector tends to be what languages use until they can write their own.