fix tool_inputs parse error in message that in CoT(ReAct) agent mode

[listeng]

Description

Occasionally, the action_input result provided by the LLM (depending on its specific capabilities, like the 'moonshot-v1-8k') may not yield a valid JSON string when operating in the CoT agent mode. This can lead to disrupt the flow of the conversation.

Fixes # (issue)

Type of Change

Please delete options that are not relevant.

• [ X ] Bug fix (non-breaking change which fixes an issue)

How Has This Been Tested?

Try to ask "计算一百二十八乘以六等于多少" with a tool eval_expression, the answer from the LLM moonshot-v1-8k will be "action_input='128乘以6的结果是768'". The fixed code will not raising an excetion in this situation, and will not interfere with other result that the LLM provided valid JSON outputs.

• TODO

Suggested Checklist:

• [ X ] I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• My changes generate no new warnings
• I ran dev/reformat(backend) and cd web && npx lint-staged(frontend) to appease the lint gods
• optional I have made corresponding changes to the documentation
• optional I have added tests that prove my fix is effective or that my feature works
• optional New and existing unit tests pass locally with my changes

[Yeuoly]

LGTM