[feature] Secret tripwire — detect & block secrets in scratches

[rwrife]

Pitch: AI coding agents and tired humans leak secrets into scratch files. Warn when a scratch looks like it holds an API key / .env-style secret, and refuse to promote it into a repo without an explicit override.

Acceptance criteria

• Heuristic scan for common secret shapes (AWS keys, *_API_KEY=, private-key headers, long high-entropy tokens)
• sp ls flags scratches that tripped the detector (e.g. a 🔑 marker)
• sp promote blocks on a tripped scratch unless --allow-secrets is passed
• sp scan <id> reports findings with line numbers (no secret values echoed in full)
• Detector lives in its own package with unit tests over fixture files

Notes

• Don't print full secret values back to the terminal — mask them.
• Keep heuristics conservative to avoid alarm fatigue.

[rwrife]

Opened #22 — implements the secret tripwire: a conservative internal/secret detector (with fixture-driven unit tests), sp scan <id> (masked findings, line numbers, non-zero exit, --json), a 🔑 marker in sp ls, and a sp promote block that refuses tripped scratches unless --allow-secrets. All acceptance criteria covered; build/vet/test green.