So a little more detail. The issue is that to use the divider for LVP, the divider voltage must be kept above about 1.8V, (brushing over the details) to prevent off-click detection. To do that you need a high voltage reference to compare the voltage pin to. A 5.0V LDO on Vcc works well for that. 3.0V doesn't work as , more on that in a bit.
So for 1S builds, instead HD reads voltage from Vcc. But in that mode it's actually reading the battery voltage from Vcc, not some regulated LDO voltage. There are only two pins where the battery voltage can be read as a higher value than the internal reference used, those are Vcc and PB0 (only Vcc presently implemented in software). So that works for 1S with no LDO.
Now back to your thing. You can use the first method, but the problem is the voltage reading stops working once the battery hits the LDO level, or possibly slightly higher (some dropout in the LDO, not sure how much). Once the battery starts dropping below the LDO the sense voltage on pin7 and the reference voltage on Vcc start to drop at the same speed and the ratio doesn't change, so you peg the bottom of the scale there.
a) So this could work, but you'll lose sensitivity somewhere above the LDO voltage and must set LVP above that. You will also probably need different resistor values for the divider.
b) One suggestion would be to instead provide battery voltage to Vcc. You only need the LDO for the voltage filter not for Vcc. However, I'm guessing the reason you aren't doing that is that you don't want to use a separate protection diode. Speaking of which, you better use an LDO with low reverse leakage. I think the ones used on the small TA boards have worked.
c) Another possibility is to find an LDO that will work at 2.5V, but there's another problem you're running up against. OTSM doesn't have as much extra headroom at low Vcc. 2.5V may not work well at all, and/or may require more attention to good hardware selection as far as the cap and the leakage of the LDO. It could work though. My specs are a bit overstated, and exactly so that things like that might just work.
d) Yet another possibility is to make a hybrid voltage readout. So that it uses both Vcc read and Vcc referenced divider read. Vcc read will remain constant until hitting the LDO voltage, then it will drop and you can get extra range from that. From a software perspective though that could be tricky, because the way bistro polls the ADC, changing references requires delays, and there's already a switch back and forth from voltage to temperature and some difficulty orchestrating the timing for that. That's not my favorite solution.
e) Yet another option is to use a separate pin with a separate voltage divider, but I'm pretty sure you have no space for that.