Add headroom_compress: 50-70% token compression for Agent Zero

[emasoudy]

Headroom Context Compression

Automatically compresses tool outputs in Agent Zero conversations using Headroom AI. Saves 50-70% tokens on every agent turn.

What It Does

• Surgical field-level compression: only compresses tool_result fields inside tool output messages
• System prompts, user instructions, behavioral rules are never touched
• Persistent stats dashboard (tokens saved, compression rate, recent activity)
• Auto-installs headroom-ai dependency via Setup button
• Works with any LLM provider

Validation

• Python syntax: all files pass py_compile
• YAML: all files pass yaml.safe_load
• Tested in production Agent Zero (Docker, GLM-5.1)
• Proven 56-73% compression on real tool outputs

Compliance

• Folder name matches ^[a-z0-9_]+$ and plugin.yaml name field
• title ≤ 50 chars (28)
• description ≤ 500 chars (286)
• index.yaml ≤ 2000 chars (469)
• tags ≤ 5 (3)
• Remote repo has plugin.yaml with matching name field
• Remote repo has LICENSE (MIT)
• Only index.yaml in submission folder