feat: Multi-modal support with automatic model fallback

- Add MultiModalManager with capability detection for vision/audio/tools
- Define fallback chains: vision (llama3.2:3b -> llava:7b -> moondream)
                       tools (llama3.1:8b-instruct -> qwen2.5:7b)
- Update CascadeRouter to detect content type and select appropriate models
- Add model pulling with automatic fallback in agent creation
- Update providers.yaml with multi-modal model configurations
- Update OllamaAdapter to use model resolution with vision support

Tests: All 96 infrastructure tests pass

src/infrastructure/models/__init__.py

@@ -0,0 +1,37 @@
+"""Infrastructure models package."""
+
+from infrastructure.models.registry import (
+    CustomModel,
+    ModelFormat,
+    ModelRegistry,
+    ModelRole,
+    model_registry,
+)
+from infrastructure.models.multimodal import (
+    ModelCapability,
+    ModelInfo,
+    MultiModalManager,
+    get_model_for_capability,
+    get_multimodal_manager,
+    model_supports_tools,
+    model_supports_vision,
+    pull_model_with_fallback,
+)
+
+__all__ = [
+    # Registry
+    "CustomModel",
+    "ModelFormat",
+    "ModelRegistry",
+    "ModelRole",
+    "model_registry",
+    # Multi-modal
+    "ModelCapability",
+    "ModelInfo",
+    "MultiModalManager",
+    "get_model_for_capability",
+    "get_multimodal_manager",
+    "model_supports_tools",
+    "model_supports_vision",
+    "pull_model_with_fallback",
+]