One naive way to do this would be to create ligature pairs for difficult pairs (mi, lm, etc). But instead, they seem to be selecting character alternates that fill the fixed width differently based on their surroundings.
I feel at least partially the same about fancy ligatures; we could just use the actual characters in source code, Unicode is widely supported. Raku does that, but it would need better typesetting (with variable-width fonts) to really shine.
It sometimes breaks out of the grid, but there still is a grid. Things are sometimes not micro-aligned but still macro-aligned, and that matters (to me).