When the Qwen Code client sent messages with content arrays like:

 1 {"role": "user", "content": [{"type": "text", "text": "hi"}]}

Pydantic parsed these as List[ContentPart] objects. The extract_text_content() function was supposed to convert these to strings, but there were edge cases where:

 1. Tool messages (role="tool") with ContentPart lists weren't being converted
 2. Harmony messages with ContentPart lists weren't being converted properly
 3. The final safety check wasn't ensuring all content was strings

 1. `_extract_text_from_content_list()`: Enhanced to handle edge cases and return empty string for empty lists
 2. `extract_text_content()`: Added ContentPart list handling for tool messages and final safety check
 3. `extract_multimodal_content()`: Added ContentPart list handling for tool messages
 4. `extract_harmony_messages()`: Added ContentPart list handling for tool and assistant messages

 - test_tool_response_message_with_content_part_list: Tests tool messages with ContentPart lists
 - Enhanced test_content_array_with_pydantic: Verifies content is string type