feat: implement Model Context Protocol (MCP) support

[Antholoj]

This PR introduces full support for the Model Context Protocol (MCP), enabling AI agents and IDEs to programmatically analyze Docker/OCI images, inspect layers, and perform deep diffing.

🚀 Key Features

• Multi-Transport Support: Implementation of stdio, sse, and the modern streamable-http transport protocols.
• MCP Toolset: New tools for agents including analyze_image, get_wasted_space, inspect_layer, and diff_layers.
• Performance: Integrated LRU caching for analysis results to minimize repeated computation.
• Docker-MCP Image: Added a specialized Dockerfile.mcp and corresponding .dockerignore for publishing a containerized MCP server to Docker Hub (antholoj/dive-mcp).
• Documentation: Comprehensive docs/MCP_GUIDE.md covering local builds, Docker usage, and client integration (Claude Desktop, Gemini-CLI, Cursor).

🛠️ Technical Notes

• Architectural Integrity: The implementation is non-breaking and strictly respects the original architectural design, utilizing the existing image, filetree, and bus internal packages.
• Go 1.24 Compatibility: Builder image updated to golang:1.24-alpine to match go.mod requirements.
• Build Optimization: Uses a dedicated Dockerfile.mcp.dockerignore to bypass root source exclusions during the MCP-specific Docker build process.
• Container Networking: Configured DIVE_MCP_HOST=0.0.0.0 by default in the MCP Dockerfile to ensure accessibility from the host.

✅ Testing

• New Tests: Comprehensive unit and integration tests for the MCP handlers and transport layers are included.
• Verification: Verified via manual smoke tests using the streamable-http transport and validated stdio mode with local AI agent integration.
• Build Validation: Confirmed full image build cycle using the new Dockerfile.mcp.

---

Disclaimer: This feature was implemented with the assistance of AI (Gemini CLI).