Keep policy engine decoupled from hook runner

[dpritchett]

Context

snag's core value is pattern-based policy enforcement across git hook phases. The hook runner (lefthook, husky, pre-commit, raw .git/hooks/) is a separate concern.

Today the policy engine (diff.go, msg.go, push.go, prepare.go, rebase.go, config.go, patterns.go) has zero lefthook awareness. This is good. The lefthook coupling is isolated to two places:

1. snag install (install_hooks.go) — writes lefthook YAML configs
2. Detection/nudge (checkout.go + shell.go) — checks for lefthook config files or .git/hooks/ scripts

Goal

Preserve this separation so snag can support other hook runners (husky, pre-commit, raw .git/hooks/) in the future without touching the policy engine.

What a multi-runner pivot would look like

• snag install becomes runner-aware: snag install --runner=lefthook vs --runner=husky vs --runner=git-hooks
• Nudge detection in checkout.go and shell hooks in shell.go learn to detect other runners
• Hint messages become dynamic ("run: snag install && lefthook install" vs "run: npx husky install")
• Policy engine stays untouched

Ground rules

• Policy commands must never import or reference a hook runner. diff.go should never know what lefthook is.
• Runner-specific code stays in install_hooks.go, checkout.go, and shell.go. If a new file needs runner awareness, that's a design smell.
• Config format (snag.toml) must remain runner-agnostic. No lefthook, husky, or pre-commit concepts leak into snag.toml.