Improve exception tracebacks for easier debugging

Summary

Make exception tracebacks clearer and easier to understand by reducing asyncio/anyio wrapper noise.

Problem

• Exceptions are often wrapped in multiple layers of asyncio/anyio task group machinery
• Tracebacks are hard to read and don't clearly point to the root cause
• The raise_exception flag is not consistently respected
• Errors from deep inside task groups obscure the actual failure point

Goal

• Cleaner tracebacks that point to user code, not SDK internals
• Raise exceptions from the main context where possible
• Reduce nesting of exception wrappers

Related

• #743 - RuntimeError but the actual root cause is not logged correctly
• #396 - MCP Server: Inconsistent Exception Handling in @app.call_tool
• #1401 - ClientSession Error Handling
• #1742 - Introduce typed error classes with metadata

_{AI Disclaimer}

该 issue 中未确定任何文件或测试。首先跟踪 asyncio/anyio 任务组路径以及 raise_exception 标志的所有使用位置，然后审查链接的异常处理相关 issue。完成的标准是：traceback 能暴露用户级别的失败，减少 wrapper 嵌套，并始终遵循 raise_exception。