I agree DynSized is desired but we also need to cope with that 2310 was postponed in favor of linting. I'd prefer reopening 2310 over unnecessary bundling.
I've discussed this in rfcs/text/0000-dyn-type.md at dyn-type-8 路 kennytm/rfcs 路 GitHub and also replied about it in Pre-eRFC: Let's fix DSTs - #2 by kennytm. Basically: you cannot override align_of_val in DynSized because the pointer is unknown, so you can only override align_of_metadata, and a thin pointer has no metadata, forcing it to be Aligned.
You could get rid of Aligned but then your Thin<dyn Trait> would be potentially unsafe to Deref.