Iterate where: explanations to fixed point

[mbovel]

When a recorded type's explanation in the where: clause mentions another recorded type that also needs explanation, the latter is silently dropped. This happens because explanations collects the entries to explain once, up front, and then renders each one; rendering an explanation for entry X may call ParamRefNameString (or similar) on entries nested inside X's bounds, which adds them to seen — but too late for the pass that produced the final output.

Concretely, for

trait Ord[T]
trait Op[A, V]
object Op:
  given Op[Int, Int] = new Op[Int, Int] {}

def min[A, V, W <: Ord[V]](x: A)(implicit op: Op[A, V], w: W): W = w
def foo(x: Double) = ???
def test = foo(min(3)) // error

we used to get:

where:    W is a type variable with constraint <: Ord[V] & Double

with no explanation for V. After this change:

where:    W is a type variable with constraint <: Ord[V] & Double
          V is a type variable

The fix is to iterate the explanation loop to a fixed point: as long as rendering an explanation causes a new entry to be recorded, collect and explain it in the next round. This also surfaces a number of previously missing explanations in capture-checking error messages (nested ^², anonymous-function parameters, etc.), which account for the bulk of the checkfile updates.

How much have you relied on LLM-based tools in this contribution?

Extensively, for everything.

How was the solution tested?

One new test showing the change, plus existing checkfiles updates.

[mbovel]

Assigning you @natsukagami because GitHub suggested so. Would you agree to review? 🙂