Why split them? I mean, the items & fields are not the only part that shouldn't be in the item tree, there're also the types, and when we remove them we get pretty much the same data - the new thing is the item tree, not the data queries.

See https://rust-lang.zulipchat.com/#narrow/channel/185405-t-compiler.2Frust-analyzer/topic/https.3A.2F.2Fgithub.2Ecom.2Frust-lang.2Frust-analyzer.2Fissues.2F7434/near/501477376

The gist is the data queries will turn into the "signatures" of the things (generic params, where clauses, generic args, etc) which for impls means generics, target trait (and its args which mat contain expressions) and the target type (again may contain expressions). All of that is information isn't necessary when querying the assoc items as they are separate concerns. Assoc item lowering is still just part of early name res, we just do it lazily because it doesn't quite participate in the fixpoint (yet, traits will be soon as rust will allow importing trait functions in the future).

The item tree is not really related to this concrete change here.

A follow up is to split out the VariantData (fields) from data queries of variants (struct_data, union_data, enum_variant_data), for similar reasons, they aren't part of the signature (same way current bodies aren't)