[Feature Request] Add --safe-mode flag to start Claude Code without user configuration

Bug Description

Title: Add --safe-mode / --vanilla flag to start Claude Code with no user configuration

Description:
There's currently no documented way to start Claude Code in a clean baseline state — without plugins, MCP servers, hooks, custom agents, slash commands, skills, or CLAUDE.md instructions loaded.

This is useful for:

• Bug reproduction — isolating whether an issue comes from Claude Code itself or from user customizations (plugins, hooks, MCPs)
• Debugging plugin/MCP conflicts — comparing behavior with vs without the user's setup
• Sharing minimal repro cases — telling someone "run claude --safe-mode in this folder" instead of "rename your ~/.claude directory temporarily"
• Testing prompts/workflows in a known-clean environment before deploying configs

Current workarounds (all awkward):

• HOME=/tmp/clean-home claude — works but breaks on Windows and loses auth
• Renaming ~/.claude/ — destructive and risky
• Manually emptying enabledPlugins in settings.json — only covers plugins, not hooks/MCPs/CLAUDE.md

Proposed:
A single flag like claude --safe-mode (or --vanilla / --no-config) that:

• Skips loading ~/.claude/settings.json and .claude/settings.json
• Disables all plugins, hooks, MCP servers, custom agents, slash commands, skills
• Ignores all CLAUDE.md / AGENTS.md files (global, project, nested)
• Keeps authentication intact (so it's actually usable)

Bonus: an opposite flag like --config-info that prints exactly what's being loaded from where, for debugging "why is this hook firing?" situations.

---

Environment Info

• Platform: win32
• Terminal: windows-terminal
• Version: 2.1.119
• Feedback ID: 536a6645-dc78-4ca1-86c2-dcd79daf0001