fix syntax error in suggesting generic constraint in trait parameter

[iximeow]

suggest where T: Foo for the first bound on a trait, then suggest
, T: Foo when the suggested bound would add to an existing set of
where clauses. where T: Foo may be the first bound if T has a
default, because we'd rather suggest

trait A<T=()> where T: Copy

than

trait A<T: Copy=()>

for legibility reasons.

the test case i added here is derived from this reproduction:

struct B<T: Copy> {
t: T
}
trait A<T = ()> {
fn returns_constrained_type(&self, t: T) -> B<T> {
B { t }
}
}

where the suggested fix,

trait A<T = ()>, T: Copy { ... }

is in fact invalid syntax!

i also found an error in the existing suggestion for trait Base<T = String>: Super<T> where rustc would suggest trait Base<T = String>: Super<T>, T: Copy, but T: Copy is the first of the trait's where clauses and should be where T: Copy as well. the test for that suggestion expects invalid syntax, and has been revised to a compiler-pleasing trait Base<T = String>: Super<T> where T: Copy.

judging by #70009 i'll.. cc @estebank ?

[rust-highfive]

r? @oli-obk

(rust_highfive has picked a reviewer for you, use r? to override)

[estebank]

Minor wording changes to comments, otherwise r=me!

[iximeow]

Good catches! tweaks applied.

[estebank]

r? @estebank
@bors r+

[bors]

📌 Commit e1607c8 has been approved by estebank