Add Gradio demo with streaming chat and image upload

[korbonits]

Summary

• app.py: interactive multimodal chat demo using gr.ChatInterface with gr.MultimodalTextbox (Gradio ≥ 4.44). Streaming output via TextIteratorStreamer in a daemon thread. Sidebar controls for temperature, max new tokens, and thinking mode. Lazy model loading on first request (thread-safe lock). CLI flags for --share, --host, --port; MODEL_PATH env var for local weights.
• README.md: new Gradio Quick Start section with install and run instructions; ticks the [ ] Online Gradio Demo roadmap item.

Features

Feature	Detail
Image upload	Single image per turn via gr.MultimodalTextbox
Streaming	Tokens streamed via TextIteratorStreamer; no waiting for full response
Thinking mode	Checkbox wires directly to enable_thinking in apply_chat_template
Temperature	Slider 0–2; 0 → greedy/deterministic
Text-only	Works without an image
Local weights	MODEL_PATH=/path/to/dir python app.py
Public share	python app.py --share for a temporary Gradio tunnel URL

Usage

pip install "gradio>=4.44"
python app.py            # http://127.0.0.1:7860
python app.py --share    # public link

Test plan

• python app.py starts without error and loads the model on first message
• Image upload + text prompt returns a streamed response
• Text-only prompt (no image) works correctly
• Thinking mode checkbox changes model output style
• Temperature=0 produces deterministic output across two identical prompts
• python app.py --share prints a public Gradio URL
• MODEL_PATH=/local/path python app.py loads local weights

🤖 Generated with Claude Code

[korbonits]

[korbonits]