docs: lead with personalized learning loop

[veltri-23]

Summary

• lead README with personalized onboarding, semantic memory, and learning over time
• position deterministic command blocking as the safe output of that learning loop
• add judge paths for onboarding, local semantic recall, returning scans, and daily strengthening
• replace stale one-scan-forever wording with bounded incremental learning

Verification

• onboarding proof: 2 pass, 0 fail
• local recall + returning service + scheduler: 20 pass, 0 fail
• git diff --check

Refs #8