fix(compilers/openapi): fill allOf position constraints

[fuad-daoud]

Summary

A value constraint written beside an allOf composing position — minProperties, maxProperties — reached a different channel than the same constraint beside a plain object schema. lowerModel filled ir.Model.Constraints from the position's own schema; lowerAllOf built its Model without doing so, leaving Constraints nil. The census that runs afterwards found no value there and concluded the keyword had no home, reporting a false openapi/degraded-construct and stashing the value under Unmodeled — while the field the diagnostic said was missing exists on every Model (#407).

lowerAllOf now reads the position's constraints with the same schemaConstraints call lowerModel makes, so the keyword lands where ir-design.md already documents it (Model.Constraints), with no diagnostic and no Unmodeled entry. buildComposedVariant, which synthesizes a Model per distributed-union variant from the same kind of position, had the same gap — and worse: that path never runs the position's census at all, so a bound written beside a co-declared oneOf/anyOf was dropped in total silence rather than misfiled. Both are fixed the same way. No other ir.Model construction site is missing the fill (the sweep is in the commit).

Test plan

• New conformance case allof-position-constraints (witnesses intersection): pins ViaAllOf.Constraints.MinProps == 2, no openapi:minProperties under Unmodeled, and no degraded-construct at /components/schemas/ViaAllOf. Planting the defect back reddens it.
• allof-oneof-cooccurrence.yaml gains minProperties: 2 on its Combo schema, since no existing fixture reached the buildComposedVariant gap (it was silent, not misfiled, so no golden could have changed); its golden now carries the bound on each distributed variant.
• make gate passes: lint 0 issues, coverage at 100%, fuzz, bench-smoke.

Closes #407

🤖 Generated with Claude Code

https://claude.ai/code/session_01TJZJ8HaugGZTjzVueafLAT