Move safety out of tablegpt-agent

[edwardzjl]

Safety should be a primary concern for application developers rather than library developers, and it is particularly not the responsibility of tablegpt-agent.

However, if we remove the safaty guard from tablegpt-agent, it will become challenging to evaluate this feature in applications like tablegpt-chat. In such cases, we may need to install the application instead of library in our eval project, again.

[edwardzjl]

@vegetablest @weekenthralling @zTaoplus @xli-1026