fix(plugin): prevent hang on PluginInvokeError via iterative unwrap

[bailong-Hakuryu]

Problem

When an Agent node invokes a plugin tool in a Chatflow, the API service hangs and eventually returns PluginInvokeError: {"message": "invocation exited without response"} after timeout.

Closes #38912

Root Cause

unwrap_plugin_daemon_error used recursive calls to peel nested PluginInvokeError payloads. A pathological or cyclic error chain could exhaust the stack and block the event loop.

Changes

• plugin_daemon_transport.py: Replace recursive unwrap with a bounded iterative loop (cap: MAX_UNWRAP_DEPTH=5, env-configurable). Fix from __future__ import annotations placement (was after import os, causing SyntaxError).'
• tool_client.py: Add raw_payload attribute to DifyPluginToolClientError and forward it through _raise_tool_daemon_error so callers retain the original daemon response body for debugging.

[sdkrre]

@bailong-Hakuryu I tried using it and added the code you provided into v1.15.0, but it still seems to hang. Is there anything else I can fix?

[bailong-Hakuryu]

我试过用它，把你提供的代码加到了v1.15.0，但它还是卡住了。还有什么我可以解决的吗？

I have updated the PR to address the underlying API hang issue.

In base_session.py, queue.Empty handling was previously entering an infinite while True: loop when waiting for MCP server responses if the receiver task remained open, ignoring read timeouts and keeping API worker greenlets blocked. An explicit max_timeout deadline check is now enforced.
Added default 60s fallback timeouts in mcp_tool/tool.py when invoking MCPClientWithAuthRetry to ensure unhandled timeouts cleanly terminate without starving Gunicorn worker connections.