You're assuming that producing a refresh card on 12nm would have cost them too much money. An optical shrink or near-shrink with tweaks would have helped tame Vega's heavy power usage near "normal" clockspeeds (1585 MHz).
AMD is selling nearly every video card they produce right now. Yes, they can keep schlepping Vega and Polaris for another year, or they can up market share a bit with 12nm Vega (and maybe even 12nm Polaris). Either way they are going to sell whatever they push out to market. The R&D expenses would likely pay for themselves. Not that we'll ever know, since AMD chose not to go that route.
So? What does that have to do with 12nm consumer cards? We already knew AMD was going to go for 7nm Vega for compute, so there was no need to address that market with 12nm parts.
I am reasonably confident that AMD has/had the resources necessary to prepare a 12nm consumer Vega and 7nm professional Vega, at the same time. They simply skipped 12nm altogether. AMD bailed out of the consumer market. I question whether or not they will ever seriously return to it.
Irrelevant to the topic at hand. Zen's success does not preclude AMD from continuing to serve the consumer dGPU market. At. All. They maintained an entirely separate RTG division under Raja Koduri while developing Zen in the first place. Continued work on Zen+ and Zen2 could have continued while RTG refined Vega and moved it to GF's 12nm process, which according to many is not much more than a refined 14nm process anyway . . .
It is not rational for them to continue selling the same consumer dGPUs for over two years.
Vega FE wasn't #1 in games when it came out last year, but AMD launched it anyway. What's stopping them now?
The feature is non-functional in Vega9, period. Whether or not developers intended to support it is another matter altogether.
AMD could likely have arranged production of HBM2 modules at GF in the event of supply shortages from Samsung and Hynix. Though I am not sure what process Samsung used for production of HBM2 last year for Vega.
AMD still managed to deliver fairly regular releases from Fiji to Polaris to Vega. Polaris obviously did not replace Fiji, but it was a new uarch that was a significant upgrade for their midrange lineup. And obviously Vega replaced Fiji. 2018 was the right time for either a 12nm Vega or small Vega, or a 12nm Polaris update at least . . . something, anything.
Nvidia is under very little pressure to update the 1070 or 1080. AMD is the only party that could apply that pressure, and obviously they are doing nothing this year.