Better workflow for creating trait impls

          I feel like a better workflow (that we should enable) would be the following maybe?:

• write impl / get a impl $0 {} completion
• start writing a trait name -> get a Trait<$1, $2, ...> for $0 completion (with each required trait arg as a tabstop), that is we now have the following state impl Trait<$1, $2, ...> for $0 {}
• then offer assists (diagnostic quickfix) for adding unresolved name segments as a new generic parameter to the surrounding impl (this one we should really have either way, its so useful)

That is, I feel like the user shouldn't even need to jump back to the generic parameter listing in the first place. If you create a generic parameter there, you will have to use it anyways and so typing out the reference to it first and then have r-a just create the definition for you sounds more ergonomic to me?

To be clear, I do agree there is an ergonomics gap here. Its just tabstop snippet completions can be a lot more annoying than helpful depending on how they are designed, and duplicating completions to accomodate for a variety of snippet setups can be more confusing than helpful which is why I (and presumably Florian) are a bit wary of this here.

Originally posted by @Veykril in https://github.com/rust-lang/rust-analyzer/issues/19447#issuecomment-2760611811

Start by tracing rust-analyzer's trait-impl completion and diagnostic quickfix paths described in the issue. Determine how to offer the impl skeleton, trait generic parameters as tabstops, and assists that add unresolved name segments as surrounding impl generics; done means this workflow works without returning to the generic parameter list.