docs: clarify when Client(raise_exceptions=True) actually raises

Description

Client exposes raise_exceptions: bool = False with a code TODO asking when raise_exceptions=True actually raises:

https://github.com/modelcontextprotocol/python-sdk/blob/main/src/mcp/client/client.py (around the raise_exceptions field)

What's missing

There is no clear user-facing documentation of:

1. Which failure modes are affected (in-process server exceptions vs transport/protocol errors)
2. How this interacts with ExceptionGroup wrapping from async with Client(...)
3. Recommended patterns vs catching MCPError inside the context manager (as docs/troubleshooting.md already teaches)

Proposed work

• Trace current behavior on main for in-memory and HTTP clients
• Document the real semantics in docs/client/ and/or docs/troubleshooting.md
• Add a small docs_src example and/or regression test if behavior is surprising
• Only change runtime behavior if docs reveal a clear bug; prefer docs-first

Related

• TODO(Marcelo) on Client.raise_exceptions in src/mcp/client/client.py
• #1401 ClientSession Error Handling (different surface; has an open PR)
• troubleshooting ExceptionGroup guidance

從 src/mcp/client/client.py 中的 raise_exceptions TODO 開始，追蹤記憶體內用戶端和 HTTP 用戶端的目前行為，包括 async with Client(...) 中對 ExceptionGroup 的處理。將其與 docs/troubleshooting.md 中的指引進行比較，然後在 docs/client/ 和/或 docs/troubleshooting.md 中記錄已確認的語義。完成標準是使用者能夠識別受影響的失敗，並在建議的模式與在 context manager 內捕捉 MCPError 之間進行選擇。