I’m inclined to agree with @Ixrec that language-level movability isn’t crucial for self-ref structs, but rather ad-hoc methods can be provided if it’s truly necessary for that type.
In either case, movability is a bit of a tangential problem for self-ref structs anyway. The real root of their unsoundness is lifetime unification problems for the field lifetimes. Until that is addressed somehow, no approach can be truly sound with resorting to HRTB closure trickery.