chore: remove the repo's dogfood failproofai hooks (#954)

The repo hooked itself: every agent CLI config in the tree ran failproofai
policies on work in this repo, through scripts/dev-hook.mjs. They block real
work, and the team has dropped them across repos. Policies now ship in
separate packs, so nothing here depends on them.

Removed the project-scope configs (.claude/settings.json, .codex/,
.cursor/, .factory/, .github/hooks/, .opencode/, .agents/hooks.json,
.agents/plugins/failproofai/, .devin/, .pi/), the .failproofai/ policies,
scripts/dev-hook.mjs, and the tests that guarded them. Product code that
installs hooks for users is unchanged.

hook-sync relied on the dogfood Stop gates to make its agent commit, push
and open the PR. The prompt now requires the agent to do that and verify
it; the entrypoint pushes and opens the PR for any commit the agent left
unpublished, and fails the run if changes were left uncommitted. The
policy-config neutralization step, the bun layer, jq and the global
failproofai install are gone.

Docs: CLAUDE.md, CONTRIBUTING.md, README.md (and its docs/i18n copies)
and .gitignore no longer describe the dogfood setup.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Author: chhhee10