feat: add Ollama embedder with batching and retries

src/companion/rag/embedder.py

@@ -0,0 +1,44 @@
+import time
+from typing import List
+
+import httpx
+
+
+class OllamaEmbedder:
+    def __init__(self, base_url: str, model: str, batch_size: int):
+        self.base_url = base_url.rstrip("/")
+        self.model = model
+        self.batch_size = batch_size
+
+    def embed(
+        self, texts: List[str], retries: int = 3, backoff: float = 1.0
+    ) -> List[List[float]]:
+        all_embeddings: List[List[float]] = []
+        url = f"{self.base_url}/api/embed"
+
+        for i in range(0, len(texts), self.batch_size):
+            batch = texts[i : i + self.batch_size]
+            last_exception: Exception | None = None
+
+            for attempt in range(retries):
+                try:
+                    with httpx.Client(timeout=300.0) as client:
+                        response = client.post(
+                            url,
+                            json={"model": self.model, "input": batch},
+                        )
+                        response.raise_for_status()
+                        data = response.json()
+                        embeddings = data["embeddings"]
+                        all_embeddings.extend(embeddings)
+                        break
+                except Exception as exc:
+                    last_exception = exc
+                    if attempt < retries - 1:
+                        time.sleep(backoff * (2**attempt))
+            else:
+                raise RuntimeError(
+                    f"Failed to embed batch after {retries} retries"
+                ) from last_exception
+
+        return all_embeddings