[User Story] - Provide support for user approval before payment gets submitted

**Description**
Payment agent ask the user to confirm the submitPayment MCP tool execution

**Current Behavior**
Today Payment Agent is instructed to ask the user to review payment details and confirm execution. this use natural language instructions that are part of the agent system prompt. This is not deterministic. We need a deterministic way to implement this human-in-the-loop pattern for real time user feedback using a conversational chat.

**Acceptance Criteria**
List clear, testable outcomes. Example:
- [ ] Payment Agent 'submitPayment' tool execution is intercepted, an approval request is sent to the chat for the user to confirm.

**Design Considerations**
- When defining payment_mcp_server in Payment Agent, use "approval_mode" params in MCPStreamableHTTPTool class. Documentation reference: [approval_mode](https://github.com/microsoft/agent-framework/blob/15433700275a2a26071016d5f824f78feb0527bb/python/packages/core/agent_framework/_mcp.py#L820C9-L820C22)
- How to configure approval_mode sample : `approval_mode={
"always_require_approval": ["processPayment"]
}`
- We need to check if there is any approval request in agent response, register the user request in memory, send an approval generated text along with request id back as part for chat response, check for approval in user response and let the agent execute the tool. An example of the user approval flow using thread: https://github.com/microsoft/agent-framework/blob/main/python/samples/getting_started/tools/ai_tool_with_approval_and_threads.py

**Main Affected Modules and/or Classes**

- app/copilot/app/agents/azure_chat/supervisor_agent.py
- app/copilot/app/agents/azure_chat/payment_agent.py
- app/copilot/app/agents/foundry/supervisor_agent.py
- app/copilot/app/agents/foundry/payment_agent.py

**Additional Context (optional)**
The current design strategy is based on agents-as-tools supervisor implementation. In the future we will migrate to the implementation based on hand-off built-in orchestration provided by agent framework and the user approval approach will probably require a significant solution refactoring.

---
*Migrated from: https://github.com/Azure-Samples/agent-openai-python-banking-assistant/issues/5*

Start with the four listed supervisor_agent.py and payment_agent.py modules, then read the MCPStreamableHTTPTool approval_mode reference and the linked approval-with-threads sample. Trace how approval requests and chat responses are represented and stored. Done means submitPayment is intercepted, an approval request and request ID reach the chat, and a user response allows the tool to execute.

Written by the indexing model from the issue text.