I think it makes sense in small doses but I think once you're using shell scripts to generate all these combinations you've lost the plot.
And I think that may be one of the larger points here. CSS can be simple, and using techniques like this forces it to stay that way. You have to take a step back and understand what we're trying to do as engineers. I would argue that being able to quickly understand code (e.g. the context to understand a line of code is small) should be a top priority of most software project.
But that doesn't matter because it's not an abstraction. If you have 10 HTML files and 1 CSS file, you can make a class called "warning" and have it be a background color red you've simplified those HTML files because you have less style information in the HTML.
But if you are using classes like "bg-red" or "pt-10" then you're not actually reducing the style content of your 10 HTML files. You've just renamed a style to a class. Sure you've reduced some possibilities but you've actually made the situation worse because now you have 11 files with style information. You've added new names for basic styles which adds even more complexity.
If you take that idea of reducing style possibilities to the logic extreme then you wouldn't be using classes like "bg-red" and "pt-10" anyway. You'd just do what everyone already does with CSS and abstract your styles appropriately.