Of course you should test in a real device, but votlage division is not a theoretical consideration nor are the guarantees on the high and low logic levels. Those are real. And a test may work fine, until you get a 1% resistor that's 1% different or a different batch of mcu's or... You need tests to prove it works, but tests don't prove it won't fail.
I definitely like considering an extra pin for power-off detection AND going back to the idea of 1.1V Vref for 1S, and maybe then just for everything. THIS can meet specs comfortably.
But even this can be thwarted by voltage division in the tailcap scenario in principle. So might as well be aware of when that starts to be an issue don't you think?
As for the tailcap, I found this:
https://budgetlightforum.com/t/-/38352
At least in that tailcap they've got resistors (DEL's R7) of 20 to 40 kohms going on. So keeping my requirement above might not be too hard. If we keep R1+R2 under 2kohmish, I guess it should work out. Of course I'm ignoring Vf, but Vf only eats voltage, which makes things better. I think it's simplest to just base it on the tailcap resistance then and figure R1+R2 should be less than 0.1 times the tailcap resistance. It could probably be 0.3, possibly even more depending on Vf, but 0.1 is a nice rule of thumb and the lower it the voltage goes, the faster it crosses threshold anyway. I think you were working in the hundreds of ohms range, so that should work.
For the rest, it still seems to me we have to use a second pin or find a way to make the adc work, or take chances playing out of bounds.
Edit: I tend to agree about the LDO being a bad idea for 1S. I'm not sure there's really a drop though or does it just fall out of regulation. I know there's a minimum drop to maintain regulation. That doesn't mean there's a minimum drop though. These are small currents. Anyway, I just don't know.