fix: make docker compose build and start the API

The image context was backend/, so COPY frontend/ failed. Build from the
repo root, place Python modules on the import path, and pin packages that
exist on PyPI. The web process no longer loads the LLM at startup.

backend/app/utils/embeddings.py

@@ -0,0 +1,19 @@
+from sentence_transformers import SentenceTransformer
+import numpy as np
+
+class EmbeddingService:
+    def __init__(self, model_name="all-MiniLM-L6-v2"):
+        self.model = SentenceTransformer(model_name)
+        self.dimension = 384
+
+    def embed(self, texts: list[str]) -> np.ndarray:
+        """Generate embeddings for a list of texts."""
+        return self.model.encode(texts, convert_to_numpy=True)
+
+    def embed_query(self, text: str) -> np.ndarray:
+        """Generate embedding for a single query."""
+        return self.model.encode([text], convert_to_numpy=True)[0]
+
+    def get_dimension(self) -> int:
+        """Return embedding dimension."""
+        return self.dimension