`SparseIR` is now super-fast to load. However, it does take quite long to precompile. I don't understand why this should be, since we mostly do bog-standard linear algebra ... or is the quad-precision linear algebra just much slower? Quite a small issue, but it is a little annoying :)