fix: surface nested target-Mac UI diagnostics

[daniele21]

Outcome

Ensure repeated target-Mac closedroom_window_missing failures expose the existing privacy-bounded AX/WindowServer diagnostic in the LOCAL REAL_ENVIRONMENT aggregate instead of losing it at the parent-runner boundary.

Root cause

real_environment_smoke.py owns the detailed UI failure and writes it to the UI evidence report. measured_release_target_mac.py and record_while_ai_busy_target_mac.py intentionally summarize that child failure as a failed check, so child_summary() was looking only at parent errors and never following ui_evidence_report / seed_ui_report.

Scope

• follow only the two canonical nested UI report references;
• require nested reports to resolve in the same evidence directory as the parent report;
• reuse the existing strict diagnostic whitelist;
• keep parent error counts and arbitrary payloads out of the aggregate;
• add regression tests for measured, contention, out-of-directory references and arbitrary private error payloads.

Four axes

• PRODUCT: PRODUCT_NONE — evidence propagation only.
• DELIVERY: INTEGRATION into dev.
• VALIDATION: repository selector authoritative; expected contained tooling scope.
• EXECUTION: REMOTE_AUTOMATED for deterministic gates; REAL_ENVIRONMENT remains required for the actual diagnostic observation.

Invariants

• no product/runtime/audio/UI behavior changes;
• no cloud fallback or external meeting-data transfer;
• no arbitrary child error payloads, UI labels/titles, transcript or meeting text copied into aggregate evidence;
• release qualification remains unchanged.