fix: make native tool calls async in async execution path to avoid bl…

[royzhz]

Problem

_handle_native_tool_calls is synchronous and uses ThreadPoolExecutor for parallel tool execution. When called from the async path (_ainvoke_loop_native_tools), it blocks the event loop, causing the entire program to hang during tool execution.

Solution

Added async counterparts for the two key methods:

• _ahandle_native_tool_calls — replaces ThreadPoolExecutor + as_completed with asyncio.gather for parallel tool execution.
• _aexecute_single_native_tool_call — tries original_tool.arun() first for true async execution; falls back to asyncio.to_thread(tool.run) if the tool hasn't implemented _arun, so the event loop is never blocked.

Updated _ainvoke_loop_native_tools to call await self._ahandle_native_tool_calls(). The sync path (_invoke_loop_native_tools) is unchanged.

Backward Compatibility

• Tools that don't implement _arun automatically fall back to thread-based execution via asyncio.to_thread — no changes required from tool authors.
• Tools that implement _arun get true async execution with no thread overhead.
• The sync execution path is completely untouched.