fix: update outdated graphify CLI commands in setup guide

[AbbassJaber]

Summary

Every graphify command shown in the setup guide is verified against graphify --help from the current graphifyy PyPI build. The CLI now uses subcommands before the path (e.g. graphify extract <path>, graphify watch <path>, graphify update <path>); the old graphify . --flag style and the --obsidian / --obsidian-dir / --mode deep flags no longer exist, so following the README as-is fails immediately.

Fixes (each cross-checked against graphify --help)

1. graphify . --obsidian --obsidian-dir <vault> → graphify extract . --out ./graphify-out. extract is the documented headless extraction command; the --obsidian* flags are not in --help.
2. graphify . --watch → graphify watch .. --help lists watch <path> as a subcommand, not a flag.
3. graphify . --update → graphify update .. Same pattern: update <path> is a subcommand.
4. Useful Commands table rewritten so every row matches a real subcommand. Added graphify explain and graphify path rows from --help. Removed the --mode deep row — that flag does not exist; semantic extraction runs automatically when an LLM API key is set.
5. Added a Step 1.5 documenting the LLM API key requirement. graphify extract errors out with "no LLM API key found" otherwise. The step also documents --no-cluster for users who want to skip LLM costs entirely.
6. graphify install → graphify install --platform claude. --help shows install [--platform P] with supported platforms claude|cursor|codex|opencode|...; bare graphify install no longer works.
7. Generated output tree updated to drop wiki/ and the auto-generated Obsidian notes folder (the current CLI emits neither). Added a note explaining how to symlink graphify-out into a vault for manual Obsidian integration.
8. Added a graphify tree step for generating GRAPH_TREE.html. The interactive visualization is no longer produced by extract — it's a separate tree subcommand per --help.

Two troubleshooting tips that referenced --mode deep and graphify update . --obsidian-dir were updated to current commands as well, so the rule "every command in the guide appears in graphify --help" holds end-to-end.

All changes mirrored in README.pt-BR.md with Portuguese translations. Conceptual sections, architecture diagrams, the Obsidian setup walkthrough, and the chat import pipeline were left untouched per the scope of this PR.

Test plan

• graphify --help lists every subcommand referenced in the updated README (install --platform, extract, update, watch, tree, query, explain, path)
• Run graphify extract . --out ./graphify-out on a sample repo and confirm graphify-out/graph.json, GRAPH_REPORT.md, and cache/ are produced
• Run graphify tree --graph ./graphify-out/graph.json --output ./graphify-out/GRAPH_TREE.html and open the HTML in a browser
• Run graphify install --platform claude and confirm ~/.claude/skills/graphify/SKILL.md exists
• Spot-check README.pt-BR.md for parity with the English changes

[lucasrosati]

Thanks for catching this — the subcommand-only CLI issue is real and well-documented
(see safishamsi/graphify#514). Most of your changes are correct fixes.

Two clarifications before merge:

1. The /graphify . style (with leading slash, no subcommand) still works when invoked
   as a skill inside Claude Code — it routes through the Python module directly. The
   subcommand style (graphify extract .) is the headless/terminal path. We should
   document both rather than replacing one with the other, since most readers of this
   guide are running inside Claude Code.

2. The --obsidian and --obsidian-dir flags do exist in the package
   (graphify.export.to_obsidian()), they're just not wired through the shell CLI parser.
   This is why the original guide worked for users running via skill but breaks for users
   running graphify . directly in the terminal. Worth mentioning explicitly.

Otherwise the LLM API key step, the graphify tree separation, and the install
platform flag are all correct catches. Could you push a follow-up commit covering
both paths (skill + headless)?