docs: add AGENTS.md universal agent brief and llms.txt files

[w7-mgfcode]

Summary

Add the AI-agent memory-file layer recommended by the emerging cross-tool standards:

1. AGENTS.md — a universal, cross-tool agent brief (the open AGENTS.md standard, read by Claude Code, Cursor, Copilot, Gemini CLI, Aider, etc.). Holds the shared project context: stack, commands, standards, testing, architecture, safety.
2. CLAUDE.md — slimmed so shared content is single-sourced: it @imports AGENTS.md and keeps only Claude-specific material (operating index, docs/_base/ imports, Safety lists, Learnings). Stays within its 150-line budget.
3. llms.txt + llms-full.txt — repo-root files following the llms.txt spec: llms.txt is an H2-sectioned navigable index of the docs tree; llms-full.txt inlines the core docs for full-context LLM consumption.

Rationale

The repo currently has only CLAUDE.md, which is Claude-only. AGENTS.md is the Linux-Foundation-backed open standard that any agent reads; adding it makes the repo's context portable across tools without duplicating instructions. llms.txt gives any LLM a clean, markup-free map of the documentation.

Out of scope / notes

• Two local .claude/skills/ skills (managing-agent-memory, managing-llms-txt) generate and maintain these files, but .claude/ is gitignored — those are local-only and not part of this PR.
• No Cursor/Copilot symlinks — single-maintainer repo; AGENTS.md only.

Acceptance

• AGENTS.md created with accurate, repo-verified commands
• CLAUDE.md @imports AGENTS.md, no duplicated content, ≤150 lines
• llms.txt + llms-full.txt follow the spec format

[w7-mgfcode]

Done in #200 (merged to dev). Closed manually — the PR targeted dev, not the default branch, so the close keyword didn't auto-fire.