chore(context7): limit Context7 indexing to the user docs

[baldurpan]

What

Adds context7.json at the repo root, ahead of submitting northguild/worktree to Context7, and records it in context/stack.md.

By default, Context7 indexes every markdown file in a repo. Here that would be about 1,900 lines of user docs against more than 11,800 lines of planning artifacts in context/, plus the agent files under .claude/, .agents/ and .github/agents/. Agents asking how to use worktree would get plan ledgers back.

What gets indexed

Context7's owner docs say root-level markdown is always indexed, and that setting excludeFiles replaces the default exclusions.

Indexed (22 files)	Excluded (119 files)
docs/src/app/docs/** (20 pages)	docs/src/app/docs/changelog/ (release policy, not usage docs)
skills/worktree/SKILL.md	context/, .claude/, .agents/, .github/agents/, .changeset/
root README.md	docs/README.md, docs/worker/README.md
	root AGENTS.md, CLAUDE.md, CHANGELOG.md (named in excludeFiles)

The counts come from running every tracked .md/.mdx file through Context7's documented rules. They are not from Context7's own parser, so check the result after submitting.

The six rules shown to agents each restate a passage of skills/worktree/SKILL.md. context/stack.md now names what has to keep this file in step: moving docs/src/app/docs/ or skills/, adding a new root .md, or changing one of the behaviours the rules describe.

No release note

context/** owes none per context/release.md. Root context7.json has no row there and isn't in the package's files list. The release-note check's path filter doesn't match either file.

Verification

• pnpm check: exit 0. This is a docs/config-only change, so per context/verify.md Gate 1 is lint plus a read of the diff. Biome doesn't cover root JSON, so the file was also checked to parse.
• Review (reviewer subagent): FAIL on the first pass, PASS after one fix. The first version of rule 3 said any remote ref skips the --source prompt, but src/commands/branch.ts:151 only exempts origin/.

After merge (maintainer steps, not in this PR)

1. Submit the repo at context7.com/add-library and check that the indexed pages match the table above.
2. Claim the library.
3. Optionally, add Context7's refresh workflow. It needs a CONTEXT7_API_KEY repository secret from the Context7 dashboard, so it was left out rather than landed as a job that fails until that secret exists.