fix(coderabbit #458): route deprecated nars_rule() through inference_type()

CodeRabbit caught that the deprecated nars_rule() shim duplicated the
Chain/ChainRev/Fork/Collider → Deduction/Induction/Abduction mapping
that already lives in the canonical inference_type() method.

Two-source mappings drift. If a future PR ever extends the junction
taxonomy or refines a mapping (e.g. Pearl figure → NARS rule discount
calibration), maintainers would have to update both sites and the v1
+ v2 APIs could disagree silently for downstream consumers.

This commit routes nars_rule() through inference_type() so the v1
shim is a transparent projection:

  match self.inference_type() {
      Some(InferenceType::Deduction) => Some(NarsRule::Deduction),
      Some(InferenceType::Induction) => Some(NarsRule::Induction),
      Some(InferenceType::Abduction) => Some(NarsRule::Abduction),
      Some(InferenceType::Revision) | Some(InferenceType::Synthesis) | None => None,
  }

Revision + Synthesis are NOT junction-derivable (no Pearl junction
maps to either), so those arms return None defensively even though
they are unreachable in practice given the Chain/ChainRev/Fork/
Collider/Unrelated exhaustion.

The existing tests (deprecated_nars_rule_matches_inference_type,
deprecated_nars_rule_none_when_unrelated, from_nars_rule_lifts_to_
inference_type) all continue to pass with the routed shim — the
observable behavior is identical; only the source of truth is
consolidated.

The CodeRabbit comment references the same learnings note that codex
P1 #457 cited: 'When a PR ships v2-feature work with v1 API backward-
compat shims, expect codex review to flag any v1 accessor that
hasn't been routed through the canonical mapping. Resolve before
merge; don\'t defer.'

Provenance: CodeRabbit review on PR #458.

Author: AdaWorldAPI

crates/lance-graph-contract/src/pearl_junction.rs

@@ -118,11 +118,20 @@ impl PearlJunction {
     )]
     #[allow(deprecated)]
     pub const fn nars_rule(self) -> Option<NarsRule> {
-        match self {
-            Self::Chain | Self::ChainRev => Some(NarsRule::Deduction),
-            Self::Fork => Some(NarsRule::Induction),
-            Self::Collider => Some(NarsRule::Abduction),
-            Self::Unrelated => None,
+        // Route the deprecated v1 surface through inference_type() so the
+        // junction → rule mapping lives in ONE place (per CodeRabbit on
+        // PR #458 — avoid the same duplication-map drift class that
+        // motivated the inference_type() introduction in #457).
+        //
+        // The full InferenceType taxonomy includes Revision + Synthesis
+        // which are NOT junction-derivable (no Pearl junction maps to
+        // either), so those arms return None defensively even though
+        // they are unreachable in practice.
+        match self.inference_type() {
+            Some(InferenceType::Deduction) => Some(NarsRule::Deduction),
+            Some(InferenceType::Induction) => Some(NarsRule::Induction),
+            Some(InferenceType::Abduction) => Some(NarsRule::Abduction),
+            Some(InferenceType::Revision) | Some(InferenceType::Synthesis) | None => None,
         }
     }
 }