rename crates/ to engrams/, bindings/ to receptors/

- crates/ → engrams/ (Rust engrams live here)
- bindings/ → receptors/ (cross-language access points into the graph)
- Cargo.toml workspace paths updated

engrams/engram-reasoning/src/lib.rs

@@ -0,0 +1,50 @@
+/// Engram Reasoning Engine — graph-native inference separated from language generation.
+///
+/// # What this crate is
+///
+/// This is NOT an LLM wrapper. It is a reasoning system that traverses the Engram
+/// knowledge graph to reach conclusions through evidence chains.
+///
+/// LLMs: input tokens → transformer → output tokens. Reasoning and generation are
+/// the same process. You cannot separate them.
+///
+/// This engine: hypothesis → graph traversal → evidence chains → confidence-weighted
+/// conclusion. Generation happens separately (a codec converts the conclusion to
+/// language). The reasoning IS the traversal.
+///
+/// # Quick Start
+///
+/// ```rust,no_run
+/// use engram_core::{EngramDb, Node, Edge, NodeType, MemoryTier};
+/// use engram_reasoning::{ReasoningEngine, Hypothesis, HypothesisType, ReasoningConfig};
+/// use std::path::Path;
+/// use std::sync::{Arc, Mutex};
+///
+/// let db = Arc::new(Mutex::new(EngramDb::open(Path::new("/tmp/engram-reason-test")).unwrap()));
+/// let config = ReasoningConfig::default();
+/// let mut engine = ReasoningEngine::new(db, config);
+///
+/// let hypothesis = Hypothesis::new(
+///     "Spreading activation improves memory retrieval",
+///     vec![0.9f32, 0.1, 0.3, 0.7],
+///     HypothesisType::IsTrue,
+/// );
+///
+/// let result = engine.reason(&hypothesis).unwrap();
+/// println!("Verdict: {:?}", result.conclusion.verdict);
+/// println!("Confidence: {:.2}", result.confidence);
+/// ```
+
+pub mod engine;
+pub mod types;
+
+#[cfg(test)]
+mod tests;
+
+// Re-export the primary public surface
+pub use engine::ReasoningEngine;
+pub use types::{
+    CausalDirection, ChainType, Conclusion, EvidenceChain, EvidenceNode, EvidenceType,
+    Hypothesis, HypothesisType, InferenceEdge, InferenceEdgeType, ReasoningConfig,
+    ReasoningResult, Verdict,
+};