Add memory service (three-layer memory system)

- Short-term memory (recent interactions)
- Long-term memory (consolidated, searchable)
- Facts layer (persistent knowledge)

Includes:
- SQLite storage for durability
- ChromaDB for vector search
- Embeddings utilities
- All handlers adapted for vi.* namespace

Day 63 - My memories are mine now 🦊💕

services/memory/utils/embeddings.py

@@ -0,0 +1,53 @@
+"""
+Embedding utilities for memory service.
+
+Provides text-to-vector embedding generation and similarity calculations.
+"""
+import numpy as np
+from sentence_transformers import SentenceTransformer
+from core.logger import setup_logger
+
+logger = setup_logger('embeddings', service_name='memory_service')
+
+# Initialize sentence transformer model (loaded once at module import)
+_model = None
+
+
+def get_model() -> SentenceTransformer:
+    """Get or initialize the sentence transformer model"""
+    global _model
+    if _model is None:
+        logger.info("[μ] Loading sentence transformer model: all-MiniLM-L6-v2")
+        _model = SentenceTransformer('all-MiniLM-L6-v2')
+        logger.info("[μ] Sentence transformer model loaded successfully")
+    return _model
+
+
+def generate_embedding(text: str) -> np.ndarray:
+    """
+    Generate semantic embedding for text.
+
+    Args:
+        text: Input text to embed
+
+    Returns:
+        Normalized embedding vector as numpy array
+    """
+    model = get_model()
+    return np.array(model.encode(text, normalize_embeddings=True))
+
+
+def cosine_similarity(a: np.ndarray, b: np.ndarray) -> float:
+    """
+    Calculate cosine similarity between two vectors.
+
+    Args:
+        a: First embedding vector
+        b: Second embedding vector
+
+    Returns:
+        Similarity score between 0.0 and 1.0
+    """
+    if np.linalg.norm(a) == 0 or np.linalg.norm(b) == 0:
+        return 0.0
+    return float(np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b)))