A couple of them seem to be behind a paywall, or login (at least for me).
I did read up a bit on the others. While I don’t doubt that memory decay begins immediately and that unquestionably short-term memory likely has a, well, shorter horizon of decay (all things being equal) I’m still somewhat hard-pressed to be convinced that, as you put it, “optimal intervals after encoding should be as short as 15-30 seconds.”
No doubt it is conceivably possible for someone to forget something in 12 seconds and, arguably, there is some benefit to all repetitions (to varying levels of effectiveness and/or efficiency), and there are always boundary cases and extreme outliers (as with anything). Maybe at the end of the day that’s just something to be kept in mind when engineering software used by large numbers of people. ¯_(ツ)_/¯
From one of the papers you shared, which I think is a general sentiment most Anki zealots are familiar with:
By this metric, surely a 15-30 interval would be rarely found as the longest possible interval that does not lead to forgetting? Perhaps not never, I guess, but I confess I shudder a bit at the thought.
This probably hedges at the conundrum I would raise: I have a hard time believing that a 1-minute (or smaller) interval could be fairly described as optimal, in almost any use-case. Again, I admit I could be wrong about this and I confess I am super interested to see the new FSRS5 intervals and fully hold space that maybe I am dead wrong and Anki will decide that I need to see things every 14.4 minutes repeatedly and into perpetuity, lolol. If so, then I resign to my fate. 
In either case, happy to be wrong and happy to learn however it might shake out.