If you do any kind of deep dive into the sensor tech you will find that the sensor does have 48MP of resolution and the question is which software output you extract. If you need the tech specs for the sensor it is an
Omnivision OV48C. 12MP is a calculated result that combines the pixels and calculates a more accurate rendering of each pixel it sends you in the DNG file. That said the 48MP mode is also a calculated output as well as is almost every digital file you get out of just about any camera you can buy. If you ever play around with building your own camera profiles (Adobe DNG Profile Editor for example) you will quickly find that no one is really giving you truly RAW sensor outputs. This is true for high end DSLRs as well where the camera manufacturer is tweaking the color and light profiles to fit their desired results. There is a very wide audience for people who want to fiddle with just about everything in sight and just as wide an audience for those who don't want to fiddle with anything at all.
If you run these images through a good Denoise filter (eg. Lightroom Denoise, DxO PureRAW, or Topaz Denoise AI) you will find that all of them clean up quite nicely (do note that DxO can't process the 70mm sensor yet). I run everything I shoot through Denoise filters whether it's from a drone or my DSLR just because I get better image quality when I do and for me image quality is everything. I did exactly the same testing you did with my old
Air 2 and discovered that the additional pixel coverage with the 48MP mode was well worth the additional processing steps. With the
Air 2 both modes truly sucked when it came to really low light anyway. The Mavic Pro 3 is far better with the 48MP images with a larger and better sensor than the
Air 2 but the Quad Bayer software techniques are borrowed from phone cameras and are basically the same.
If your use case is using the captures with a minimum of processing if any processing then you might well be better off with the 12MP mode. I never use my images that way and have lots of techniques for reducing noise and enhancing detail so for my use case 48MP wins out hands down. I personally don't see any reason why anyone's approach is either right or wrong and it all boils down to what YOU want out of YOUR equipment and how you want to use it.
Keep playing with the two modes and include some real world imaging with your own favored processing techniques and chances are you will settle on one vs. the other as it works for how you wish to use your 70mm lens/sensor system. You may even find use cases for both modes. The best part about digital imaging vs. film is that taking extra images with digital is free and you can always delete what you don't want or didn't work.