Fix the Ollama FIM completion

We were not passing the suffix at all which was breaking the FIM
completion.

Author: jhrozek

src/codegate/providers/ollama/completion_handler.py

@@ -89,9 +89,20 @@ async def execute_completion(
         """Stream response directly from Ollama API."""
         self.base_tool = base_tool
         if is_fim_request:
-            prompt = request["messages"][0].get("content", "")
+            prompt = ""
+            for i in reversed(range(len(request["messages"]))):
+                if request["messages"][i]["role"] == "user":
+                    prompt = request["messages"][i]["content"]  # type: ignore
+                    break
+            if not prompt:
+                raise ValueError("No user message found in FIM request")
+
             response = await self.client.generate(
-                model=request["model"], prompt=prompt, stream=stream, options=request["options"]  # type: ignore
+                model=request["model"],
+                prompt=prompt,
+                suffix=request.get("suffix", ""),
+                stream=stream,
+                options=request["options"]  # type: ignore
             )
         else:
             response = await self.client.chat(