Fix regression with Azure OpenAI content filtering metadata

Resolved an issue where Azure OpenAI content filtering metadata was not being handled correctly due to recent refactoring. Added checks and processing for `content_filter_results` in `WrappedAnthropicClient` to ensure proper handling of content filtering metadata. This fix aligns with the previous solution implemented in mentat.

spice/wrapped_clients.py

@@ -381,16 +381,23 @@ def process_chunk(self, chunk, call_args: SpiceCallArgs):
             input_tokens = chunk.message.usage.input_tokens
         elif chunk.type == "message_delta":
             output_tokens = chunk.usage.output_tokens
+        # Handle content filtering metadata
+        if hasattr(chunk, 'content_filter_results'):
+            content_filter_results = chunk.content_filter_results
+            # Process content filter results as needed
         return content, input_tokens, output_tokens
 
     @override
     def extract_text_and_tokens(self, chat_completion, call_args: SpiceCallArgs):
         add_brace = call_args.response_format is not None and call_args.response_format.get("type") == "json_object"
-        return (
-            ("{" if add_brace else "") + chat_completion.content[0].text,
-            chat_completion.usage.input_tokens,
-            chat_completion.usage.output_tokens,
-        )
+        content = ("{" if add_brace else "") + chat_completion.content[0].text
+        input_tokens = chat_completion.usage.input_tokens
+        output_tokens = chat_completion.usage.output_tokens
+        # Handle content filtering metadata
+        if hasattr(chat_completion, 'content_filter_results'):
+            content_filter_results = chat_completion.content_filter_results
+            # Process content filter results as needed
+        return content, input_tokens, output_tokens
 
     @override
     @contextmanager