Apply false-green hardening to Uncle Bob harness plus

[tninja]

Summary

Apply the false-green hardening merged in #489 to both Uncle Bob's Coding Agent Harness+ prompt variants.

Harness+ shares the same evidence-first core as the regular Uncle Bob harness, but adds maintainability sensors, dependency-boundary checks, semantic modularity review, and final sensor reporting. This PR updates only that shared core and preserves the Plus-specific layers and evidence sections.

Changes

• treat the gauntlet itself as software that can produce false greens
• clarify spec re-approval after clarification changes and persist the approved spec when practical
• require gauntlet layers to be executable gates, not passive reports
• make changed-line coverage fail when its requirement is missed
• prefer established mutation tools and require manual mutation to prove each mutant was applied and executed
• make custom/home-grown gates fail closed and require a known-bad negative control
• classify unrun layers as N/A, UNAVAILABLE, or SUBSTITUTED
• require orchestration to verify declared layers actually executed and enforced their constraints
• carry over isolated-worktree safety and reproducible gauntlet entry-point hardening
• update the Evidence Report template to use the new layer classifications while retaining Sensor Status and Maintainability Review

Harness+ behavior preserved

This intentionally keeps the existing Plus-only behavior unchanged:

• dependency-boundary layer
• semantic modularity review
• maintainability sensor guidance and final sensor gate
• Tier 2/Tier 3 modularity calibration
• independent modularity pass for material architecture risk
• sensor status and maintainability review in EVIDENCE
• diagnostics baseline/get_diagnostics constraint in the diagnostics variant

Scope

Prompt-only change. No Elisp runtime behavior is modified.

Follow-up to #489.