There are two objectives behind my suggestions for modifications. Firstly to avoid the need to perform surgery on the Pyboard, and secondly to avoid the need for external components. The first is fixed by the new regulator.
If you provide a switched 3.3V supply, the pullup problem goes away: you merely run the pullups from that supply and ensure in firmware or otherwise that the 3.3V supply is normally on (see below).
would it switch the existing 3V3 pins
I can't see a reason why not. The only one which comes to mind might be if someone wanted to use the pins to source power to the Pyboard but, as I suggested in my notes, I see little point in doing this.
This is something that's easy to add yourself if needed,
For us, but some users lack this knowledge of hardware. This could be addressed with clear instructions.
do you think it would be widely used
The $64000 question. How can we possibly assess this? My gut feeling is that ultra low power consumption is an amazing feature of the Pyboard but it's very possible that 99% of boards get powered only by USB.
The lack of an I/O pin could kill the idea. One idea might be to bring the gate of the MOSFET out to an extra pin. A solder bridge would link the gate to Gnd. If the user wanted to employ the switch, they would remove the bridge and link the gate to a pin of their choice. I'm perhaps showing my age with the idea of a solder bridge: are they feasible with automated production?
With the bridge in place DFU mode would work normally. Users removing the bridge would need to be aware that the pin should be linked to Gnd before entering DFU.
Regarding the I2C pullups where we don't have an onboard MOSFET: the user is then committed to adding external components so two resistors won't be a significant addition. Your solution of an alternative I2C bus without pullups seems good and removes the need for the analog switch required by the existing design.
