I noticed the RAK19007 replaced the TPS27082L P-FET switch for 3v3_S with a P-FET driven by a bipolar inverter. As a result, the ON-state quiescent current went from ~1.2uA to ~1.1mA, problematic in an application where the 3v3_S rail is used for external-interrupt pull-ups (a completely reasonable thing to do) with the MCU in sleep.
From what I can tell, only the P-FET is required, since the MCU core drives IO2 with a 3v3 signal. A 2M pull-up is more than sufficient to keep 3v3_S off during MCU reset. This does mean IO2 would invert where IO2 = 0 turns on 3v3_S, though. ON-state quiescent current would be under 2uA.
Alternatively, substantially increasing R20 from 4k7 to, say, 2M, and replacing Q8 with, say, a DTC with 100k base resistor, could get quiescent ON-state current under 40uA. Substantially better but not as good as the TPS27082L.
Requests for a future rev perhaps.