feat: review agentic workflows end-to-end with multi-agent validation

[Sealjay]

Context

The agentic layer (AI Dungeon Master, combat agents, narrative generation, world building) needs a comprehensive end-to-end review to ensure all agents work correctly, are accurately configured in Azure AI Foundry, and the data layer supports the full workflow.

Tasks

• UI Agent review: Verify all frontend surfaces correctly trigger and display agent responses
• UX Agent review: Ensure agent interactions feel natural, errors are handled gracefully, loading states are clear
• Agent Specialist review: Validate all agents (DM, combat, narrative, world gen) are correctly configured in Foundry with appropriate models, system prompts, and tool registrations
• Service Designer review: Map the full user journey through agent interactions — identify gaps, dead ends, or broken flows
• Devil's Advocate review: Stress-test edge cases — what happens when AI is unavailable? When fallback mode kicks in? When multiple agents conflict?
• Verify all agents work with the singleton azure_openai_client and handle is_configured() == False
• Confirm the data layer persists all agent state (threads, sessions, combat state, narrative arcs)
• Validate that the entire game could theoretically be played via agent API calls alone (CRM-style agentic access)
• Test circuit breaker behaviour under agent failures

Why

The agentic layer is the core differentiator of this project. A multi-perspective review ensures it works reliably and provides a good user experience across all interaction modes.