docs: document session-command skills and code-location guidance (EN + pt-BR)

[shiyao821]

I mostly added these as I had some misconceptions while reading the README.md. Making these points explicit I think will help the next person.

What

Updates the setup guide (both English and Portuguese READMEs) with three additions:

• ⚠️ "Keep code out of the vault" warning — clarifies the vault holds notes only, and that code repos must live elsewhere so Obsidian doesn't choke on node_modules/, build output, etc.
• New setup step: turning /save + /resume into global skills — explains implementing the session commands as skills under ~/.claude/skills/ so they fire reliably in any project, instead of repeating the prose in every CLAUDE.md.
• "Unknown Command" troubleshooting entry — explains the headless-CLI-vs-skill distinction when graphify . errors with unknown command '.'.

Plus a minor note that graph.json / GRAPH_REPORT.md are worth keeping versioned for the team.

Why

The skills-based approach is more reliable than CLAUDE.md prose, and new users were putting code repos inside the vault (killing Obsidian performance). The two changes address the most common setup mistakes.

Notes

README.pt-BR.md mirrors README.md, adapted to the pt-BR conventions already in that file (/retomar, /salvar, projeto/).
Docs-only change — no code affected.

[lucasrosati]

Great additions — the vault warning and the skills section address real confusion points (see #5).

Two things before merge:

1. The "Unknown Command" troubleshooting entry only made it into README.pt-BR.md —
   the English README.md diff doesn't include it. Could you mirror it there?

2. In that troubleshooting block, --mode deep is listed among the skill-form flags,
   but we removed that flag in the recent CLI-accuracy PR (semantic extraction now runs
   automatically when an LLM API key is set). Please drop it from the example to keep
   the guide consistent.

Happy to merge once those two are in.