In cloud resolver, encourage the agent to ask for help when task is unachievable

[neubig]

Currently in the cloud resolver, the agent is instructed to finish the task, and it will try to do so at all cost. However, in many cases the task is not achievable and we should instead ask the user for clarification.

We can reference this research paper (by @sani903): https://arxiv.org/abs/2502.13069
And also I believe that @xingyaoww is working on a way to determine whether issues are solveable or not.

CC @malhotra5 as well.