compilers/openapi: a boolean allOf branch composes an empty open model

[OmarAlJarrah]

A boolean schema used as an allOf branch contributes nothing, and the composed model says the
opposite of what the source does.

components:
  schemas:
    Never:  {allOf: [false]}
    Always: {allOf: [true]}

Both lower to kind: model with no properties, no additional, nothing under unmodeled, and no
diagnostic. An empty open model admits every instance — so Never, whose source admits none, is
lowered to the most permissive shape the IR has.

Why this is inconsistent rather than merely incomplete

ir-design.md §4.8 already states the handling for a boolean schema in its own right: "a boolean
false schema becomes a closed empty Model with an info one". The composition path does not
apply that rule to a branch, so the same source construct is lowered two different ways depending
on where it appears, and only one of them is documented.

Why it was left out of the branch-residue work

The residue derivation keeps whatever an inline branch declares that the merge does not consume,
derived from the branch key set. A boolean branch declares no keys at all, so there is nothing for
that mechanism to rescue — the gap is not preservation but lowering: what should the composed node
be when one conjunct admits nothing. That is a decision worth making deliberately rather than
falling out of an empty merge, which is why it was recorded in the code and the PR body rather than
patched.

Suggested shape

Apply §4.8s existing rule through composition: a false branch makes the composition unsatisfiable,
so the composed model should be the closed empty Model with the same info diagnostic. A true
branch is a genuine no-op and can stay silent, though saying so once would cost little.

Method

Compiled through openapi.New().Compile(...), inspecting the emitted models and the full
diagnostic list.