Investigate adding an AgentFinishReason abstraction

`ChatFinishReason` as part of `ChatResponse`, and is being added to `AgentRunResponse`. The purpose of this task is to investigate if we need to extend the possible finishreasons for agents, for instance for user approvals or other reasons why a Agent would stop vs a ChatClient. We should look at agents that are not wrappers for LLM's such A2A, Copilot Studio, ...

Start by reading ChatFinishReason as used by ChatResponse and AgentRunResponse, then compare agents that are not wrappers for LLMs, including A2A and Copilot Studio. Determine whether agent-specific reasons such as user approvals are needed and document the resulting abstraction and completion criteria.

Written by the indexing model from the issue text.