Keep full conversation context across HITL multi-round runs

Resolved a bug where add_history_to_context was not correctly applied during Human-in-the-Loop runs that involved multiple conversation rounds. Agents paused for human review and subsequently resumed now have access to the full conversation history in context, preventing gaps in reasoning across approval boundaries.

‍

Details:

Fixes history injection for HITL workflows using add_history_to_context across multiple rounds
Ensures agents resuming after a pause have full conversational context available
No configuration changes required; the fix applies automatically to existing HITL setups