It's a bit confusing indeed, and to make it worse there are some differences between regions within the same generation too. But to keep things simple, all generations do have decent "shared" DDoS protection while G1 also offers additional 10-100GBit/s protection from within the region, this reduces the latency compared to using external services for DDoS protection, which of course is also a possibility if needed.
I don't know about dropped traffic or null routing, never had that issue. But I don't think my sites are very popular target for DDoS attacks either. In any case, if you're concerned about DDoS attacks you might wanna stay away from G2 servers as G1 and G3 does in fact provide better DDoS mitigation out of the box, while G1 also offers additional protection for an additional fee.
Memory optimized, CPU optimized and that sort of things indicates the hardware allocation, Compute and dedicated instances for instance offer higher CPU frequencies, while memory optimized provides more memory per CPU core, 8GB/core I think. A bit confusing but memory cant be shared anyway so it has to be optimized in additional ways.
Standard and optimized are just cheap as far as I know, best suited for development but works pretty good even for smaller production environments. Optimized says "none" on included bandwidth but they don't charge for bandwidth currently, other than that it's a lot of server for your money if your bandwidth usage is low. And gpu instances comes with a GPU attached to it I guess. 😉