QSA exact typed Horn logic closure

[Inserian]

Adds an exact bounded Horn-style implication engine as the next typed QSA mathematical-language primitive above accepted PR78.

The child consumes only caller-admitted opaque proposition identities, signed facts and typed implication rules. It does not parse raw language, decide semantic truth, or select final answers. It provides:

• explicit three-state query semantics: True, False, Unknown;
• signed literals, including exact negative premises/conclusions;
• conjunction-premise Horn rules with canonical duplicate removal;
• linear watcher/count forward closure over premise incidences and newly derived literals;
• exact chained implication and entailment;
• contradiction detection with rejection before accepted facts/rules are mutated;
• contradictory-premise rejection, duplicate fact/rule no-ops and explicit resource caps;
• insertion-order-independent canonical identity over atoms, admitted facts/rules and exact closure;
• QMath language route HornLogic with Boolean type and ExactStructural evidence.

A dedicated native gate checks conjunction, chained implication, negative premises, true/false/unknown queries, receipt accounting, contradiction nonmutation, duplicate suppression, contradictory premises, insertion-order-independent canonical identity, QMath route/type/evidence preservation and caps.

Brain boundary: Brain owns language interpretation, proposition identity, truth admission, procedure binding, memory and final realization. QSA receives only admitted logical facts/rules and performs exact closure/entailment with receipts. This is intended as a structured substrate for Brain logical reasoning, not a natural-language rule parser or unrestricted SAT theorem prover.

Research-only stacked successor to accepted PR78; no merge or production promotion is authorized.