I'm a software developer working at Felt building the future of mapmaking software, and I used to work on the X-Plane flight simulator. I work primarily in Elixir & React, but I've written a lot of C++, Swift, Elm, and Python as well.
The biggest thing missing from this analysis is "is there a business case for supporting those 2% of users?". (Maybe, maybe not.)
The second biggest thing is progressive enhancement. The author picked a CSS feature (nesting) that is basically all-or-nothing: the site will basically be entirely broken for those 2% if you swap Sass for native nesting. Most features aren't like that; maybe the site won't look pixel-perfect on old browsers, or one bit of functionality won't work, but by and large it will still be functional. In those cases, I think it's a much easier decision in terms of where to draw the cutoff.
Having been a happy customer of Blacksmith, Depot, and Ubicloud at various points (all three of which offer GitHub Actions runners at cheaper prices than GitHub themselves), I agree the billing here is weird. It's worth pointing out, though, that you have to use a lot of CI minutes to incur a $1000 bill. You're well outside of hobbyist levels, and that same level of usage would have previously cost double from GitHub themselves. This a business with serious compute needs, being asked to pay real money.