More pixels in low light could allow you to do more intelligent processing, maybe even in a way that allows for longer exposures or multiple exposures to do the job of gathering more light (without incurring as much of a fidelity penalty as usual from motion blur.)

More pixels provide more details, but not always more quality, as poor lens quality is also "enhanced". This is useful for large prints which require a high dpi. In low light, less pixels in a greater sensor area perform better. Dense pixels result in more iso noise

Given the same sensor area, a dramatic increase in pixels will reduce low-light sensitivity. The space between pixels does not collect light, so the more spaces between, the less light collected overall.

Think of a tic-tac-toe grid - it's mostly white space with a few black lines. Expand that out to a 1000x1000 grid in the same space, and there is a lot less white space and a lot more black lines.

We have been using gapless microlenses on imaging sensors for quite a while now. Also, when the pixels shrink, tech advancements in reducing downstream read-noise also mean that very little ISO performance is sacrificed (if any). Certainly, there is a lot more nuance and tech to discuss, but those are the broad strokes.

https://www.photonstophotos.net/Charts/PDR.htm#Canon%20EOS%2...

https://www.photonstophotos.net/Charts/RN_e.htm#Canon%20EOS%...

Diffraction blurs the image even for 10MP, given the small lens/sensor. The other 98 MP are used to remove noise, especially at low light

I assume this is similar to delta-sigma ADC/DAC, where you shift all the noise to a higher spectrum, so you can easily cut it all off?

I dunno, I took optics 15 years ago and only one signals class :P

I'd just average the lowest (N-1) pixels. N-1 because shot noise is a very bright pixel - if I have a bright pixel not including it is best for the avg. If the pixel is not surging, not including it doesn't affect my average much.