Add evidence contract and agentic evaluation loop

[MicroMilo]

What changed

• Added an explicit report contract (schema_version 1.1) with measurement status, coverage ledger, evidence-linked findings, confidence, and an action gate.
• Changed incomplete size probes to remain null with timeout/error/unknown status instead of being interpreted as zero.
• Marked artifact parent relationships and whether artifact bytes are already included in a measured target.
• Added validate_report.py, deterministic timeout contract tests, and CI coverage.
• Added a maintainer agentic evaluation loop reference using aggregated redacted evidence packets and independent role reviews.
• Updated the README, skill instructions, schema reference, and synthetic report example.

Why

A real local baseline showed a timed-out Desktop probe, 251 Git repositories with dirty work, 40 artifact rows, and app-managed storage. The previous report could prioritize review but did not preserve enough evidence state for another Agent to safely reason about exact actions.

Evaluation

• Ran a real local deep audit with no cleanup or file mutation.
• Used independent audit-operator, first-time-user, and loop-architect subagents against an aggregated redacted packet.
• Recorded the privacy-guardian role as incomplete after timeout; it was not treated as approval.
• python3 tests/smoke_test.py
• python3 tests/contract_test.py
• python3 audit-local-files/scripts/validate_report.py on the real local report
• skill-creator quick_validate.py passed

Raw local reports and role packets are not included.