docs: make first tasks and installation feedback reproducible

[morningstarnasser]

What changes

• Add three small first-task recipes with independent acceptance checks and explicit model-cost / demo-evidence limits.
• Use single-line prompts for the terminal's one-line-per-turn input.
• Link first tasks and first-install feedback from the README and contribution guide.
• Add a structured first-run issue form with a required privacy review and private security-reporting route.
• Add three offline proof groups reusing 13 existing regression cases, including task approval and no approval carryover to the next task; run them in CI.

No runtime, policy, provider, dependency, release, version or website deployment changes.
No model-backed demonstration is claimed by the offline fixtures. No fabricated installs,
testimonials, rankings or comparative speed claims.

Verification

• Three fixture groups twice: 13/13 each run.
• Full pytest: 3217 passed, 2 skipped (3219 collected).
• Redteam: 263/263; optional installed-model-worker credential probe skipped because this development checkout has no such worker environment.
• Form YAML structure, required privacy checkbox, all CLI cases and invalid-case rejection checked.
• Three single-turn prompts, local links and FAQ fragment checked.
• FAQ discovery checks, public hygiene and git diff --check: passed.
• Gitleaks staged diff: no leaks.
• OSV all five pinned lockfiles (50 package records): no issues found.

External first-install feedback and model-backed recipes remain to be collected; these
changes make that work reproducible, they do not pretend it has happened.