Rollup merge of #120719 - compiler-errors:no-dyn-atb, r=lcnr

Remove support for `associated_type_bound` nested in `dyn` types

These necessarily desugar to `impl Trait`, which is inconsistent with the `associated_type_bound` feature after #120584.

This PR keeps the `is_in_dyn_type` hack, which kind of makes me sad. Ideally, we'd be validating that no object types have associated type bounds somewhere else. Unfortunately, we can't do this later during astconv (i think?), nor can we do it earlier during ast validation (i think?) because of the feature gating of ATB being a *warning* rather than an *error*. Let me know if you have thoughts about this.

r? lcnr