Refactor LLM processor and improve async handling

Move contextual information handling from noise filtering to extraction
and centralize LLM call logic. Wrap blocking transcription and state
update calls in asyncio.to_thread to prevent event loop blocking.
Update transcriber model size to base.

src/llm/processor.py

@@ -61,6 +61,18 @@ class LLMProcessor:
 
         self.model = model or os.environ.get("LLM_MODEL", "gpt-4o")
 
+    def _strip_markdown_code_blocks(self, content: str) -> str:
+        """
+        Strips markdown code blocks (e.g., ```json ... ```) from the content.
+        """
+        import re
+
+        # Remove opening and closing code blocks
+        content = re.sub(
+            r"^```(?:json)?\n?|```$", "", content, flags=re.MULTILINE
+        ).strip()
+        return content
+
     def _call_llm(
         self,
         system_prompt: str,
@@ -93,15 +105,7 @@ class LLMProcessor:
             )
             content = response.choices[0].message.content
 
-            # Strip markdown code blocks if present
-            if content.startswith("```"):
-                import re
-
-                content = re.sub(
-                    r"^```(?:json)?\n?|```$", "", content, flags=re.MULTILINE
-                ).strip()
-
-            return content
+            return self._strip_markdown_code_blocks(content)
         except Exception as e:
             logger.error(f"LLM Error: {e}")
             return ""
@@ -147,34 +151,19 @@ class LLMProcessor:
         """
         logger.info(f"LLM Processor (Extract): Calling extraction for: {filtered_text}")
         try:
-            # Using standard chat.completions.create with JSON mode for better compatibility with vLLM
-            logger.info("LLM Processor (Extract): Sending request to backend...")
-
             system_prompt = EXTRACTION_SYSTEM_PROMPT
             if context:
                 system_prompt += f"\n{context}"
 
-            messages = [
-                {"role": "system", "content": system_prompt},
-            ]
-            messages.append({"role": "user", "content": filtered_text})
-
-            for message in messages:
-                logger.info(f"LLM Processor (Extract): Message: {message}")
-
-            response = self.client.chat.completions.create(
-                model=self.model,
-                messages=messages,
+            result = self._call_llm(
+                system_prompt=system_prompt,
+                user_prompt=filtered_text,
                 response_format={"type": "json_object"},
-                extra_body={"enable_thinking": False},
             )
-            logger.info("LLM Processor (Extract): Response received from backend.")
 
             import json
 
-            content = response.choices[0].message.content
-            logger.info(f"LLM Processor (Extract): Raw JSON response: {content}")
-            data = json.loads(content)
+            data = json.loads(result)
 
             # Map the JSON data to the Pydantic model
             return ExtractionResult(**data)