The only reason people are referring to the base model as a downgrade is because it no longer has a dGPU. Here's the thing, though: Iris Pro is at most 20% slower than the 650m (and the gap is closer in some situations, with Iris Pro sometimes winning in compute tasks). Unless you're a serious gamer (or need the best possible 3D performance for some other application) this is not even worth thinking about.
As far as CPUs, the late 2013 model is faster than the early-2013 model at each tier. It also should use less power (on average) to deliver that performance due to the Haswell low-load optimizations.
You absolutely do not need 16GB for your usage (even 8 is probably overkill, but it gives you headroom to do things like leave 40 tabs open if you want). And in general don't think too hard about 5-6 years down the road, because the reality is that tech moves so fast that you'll be wanting to sell and replace before then regardless of how aggressively you upgrade this machine.
Example: I have a desktop I built about 5 years ago now (give or take... it was a Q9650 if you want to pinpoint when that was released). I've replaced it since, but it was really top of the line for the time: That CPU, a GTX 285, 8GB RAM. Let's check out some Geekbench scores:
http://browser.primatelabs.com/geekbench3/search?utf8=✓&q=q9650
This thing gets its butt handed to it by the lowest tier MBA. From 2011. Just something to think about.