feat(rag): embedding versioning (#6) + chunking YAML config (#8)

Embedding: EMBED_MODEL_NAME/VERSION/VERSION_STAMP/BUILD_ID env vars.
Chunking: chunking_config.yaml with 8 content types + strategies.

app/rag/chunking_config.yaml

@@ -0,0 +1,33 @@
+rag_chunking:
+  news:
+    strategy: recursive
+    chunk_size: 512
+    overlap: 0.15
+  scam_report:
+    strategy: semantic
+    chunk_size: 400
+    overlap: 0.10
+  contract_code:
+    strategy: code
+    chunk_size: 256
+    overlap: 0.20
+  wallet_profile:
+    strategy: fixed
+    chunk_size: 128
+    overlap: 0.0
+  annual_report:
+    strategy: recursive
+    chunk_size: 800
+    overlap: 0.15
+  social_post:
+    strategy: fixed
+    chunk_size: 256
+    overlap: 0.05
+  market_data:
+    strategy: semantic
+    chunk_size: 300
+    overlap: 0.10
+  default:
+    strategy: recursive
+    chunk_size: 512
+    overlap: 0.15

app/rag/embeddings.py

@@ -29,6 +29,13 @@ log = logging.getLogger(__name__)
 
 EMBEDDING_DIM: Final = 1024
 
+# Embedding versioning (ADR-0007 compliance)
+# Stamped on every embedding so downstream consumers can migrate
+# when the model changes. Format: {model}:{version}:{dim}
+EMBED_MODEL_NAME = os.getenv("RAG_EMBED_MODEL", "bge-m3")
+EMBED_MODEL_VERSION = os.getenv("RAG_EMBED_VERSION", "1.0")
+EMBED_VERSION_STAMP = f"{EMBED_MODEL_NAME}:{EMBED_MODEL_VERSION}:{EMBEDDING_DIM}"
+EMBED_BUILD_ID = os.getenv("RAG_EMBED_BUILD_ID", "")  # git commit hash or CI build number
 # Backends
 _BACKEND = os.getenv("RAG_EMBED_BACKEND", "auto").lower()
 OLLAMA_URL = os.getenv("OLLAMA_URL", "http://host.docker.internal:11434").rstrip("/")