Account for embedding depth shadowing

[mdempsky]

Consider:

package x

type A struct{ B }
type B struct{ F int }
type T struct { A; B struct { F int } }

var _ = T{}.F

Invoking unbed on A.B currently rewrites T{}.F to T{}.B.F, but the explicit B selection matches against T's field because it has a shallower embedding depth than T{}.A.B. This is a change in semantics.

Unbed could rewrite to T{}.A.B, but in general there are multiple possible qualifications that could be valid, and it's not obvious how unbed should choose one.

In the mean time, unbed should at least detect and reject this scenario.