structuredContent is dropped from isError=True tool results (but preserved on success)

## Summary `_convert_call_tool_result` extracts `CallToolResult.structuredContent` into the `MCPToolArtifact` only on the success path. When `isError=True` it raises `_MCPToolExecutionError` **before** `structuredContent` is read, so a machine-readable error payload is silently lost. The resulting `ToolMessage(status="error")` has `artifact=None`, and if the error carried no text block the model only sees the adapter's placeholder string. The same field is preserved when the identical tool returns `isError=False`. ## Versions - langchain-mcp-adapters 0.3.2 - mcp 1.29.1 - langchain-core 1.6.1 - langgraph 1.2.11 - Python 3.14 ## Reproduction Self-contained, no network / no real server (uses `mcp`'s in-memory session): ```python import anyio import mcp.types as types from mcp.server.lowlevel import Server from mcp.shared.memory import create_connected_server_and_client_session from langchain_mcp_adapters.tools import load_mcp_tools server = Server("repro") STRUCTURED_ERROR = {"code": "RATE_LIMITED", "retry_after": 30} @server.list_tools() async def list_tools() -> list[types.Tool]: schema = {"type": "object", "properties": {}} return [ types.Tool(name="ok", description="success + structuredContent", inputSchema=schema), types.Tool(name="err_structured_only", description="error, structuredContent, no text", inputSchema=schema), types.Tool(name="err_structured_and_text", description="error, structuredContent + text", inputSchema=schema), ] @server.call_tool() async def call_tool(name: str, arguments: dict): if name == "ok": return types.CallToolResult( content=[types.TextContent(type="text", text="done")], structuredContent={"rows": 3, "status": "ok"}, isError=False, ) if name == "err_structured_only": return types.CallToolResult( content=[], structuredContent=STRUCTURED_ERROR, isError=True, ) if name == "err_structured_and_text": return types.CallT…