feat(agent): conversational agent workflow control with transparent proposals (M9)

[BerryUIKI]

Summary

This PR implements Milestone M9: Conversational Agent Control of Existing Workflows and Engines.

Key Capabilities

• Natural Language Intent Parsing: The backend \AgentService\ parses natural language requests across text-to-image, image-to-image, inpainting, upscaling, text-to-video, and image-to-video actions, with automatic extraction of aspect ratios, quality/step count, random seeds, and engine preferences.
• Compound Multi-Step Action Plans: Detects chained intents (e.g. 'generate an image in 16:9 then upscale 2x') and generates multi-step execution pipelines.
• Transparent Human-in-the-Loop Gate: Formulates structured \AgentProposal\ objects that explicitly disclose target engine, model checkpoint, parameters, number of calls, and cost/privacy disclaimers (local private inference vs. cloud BYOK billing). Execution is strictly blocked until the user reviews and clicks Approve & Run.
• Frontend Conversational Panel: Slide-over drawer with chat message thread, quick prompt chips, active canvas image context indicators, transparent proposal cards, and direct canvas node instantiation on completion.

Automated Verification

• Backend pytest: 89 passed (including 6 new tests in \�ackend/tests/test_m9_agent.py)
• Frontend: TypeScript typecheck and Vite build passed cleanly
• Launcher: \cargo test\ passed cleanly