Clean up legacy record surfaces after architecture refactors

[Xunzhuo]

Goal

Clean up legacy record/code paths left from repeated architecture refactors, and make the current Personal-Model-first architecture easier to reason about and maintain.

Background

The project has gone through several major refactors. The canonical direction is now the Understanding System: Personal Model facts/questions, Elephant State, Episode/Loop/Step trail, contextual recall, and background learning jobs. Legacy naming, compatibility shims, stale records, and obsolete code paths increase agent confusion and long-term maintenance cost.

Scope

• Audit for legacy record/memory/evidence/grounding/learning surfaces that no longer match docs/system-design/system-layer-model.md.
• Remove dead code, stale compatibility paths, unused tables/methods, and obsolete tests where safe.
• Rename or document remaining compatibility shims that must stay temporarily.
• Add tech-debt entries for any divergence that cannot be removed in the same work.
• Strengthen tests around canonical Step/Facts/Questions paths after cleanup.

Acceptance Criteria

• A legacy-surface inventory is produced before removal work starts.
• Removed surfaces are covered by tests or explicit no-longer-referenced proof.
• Any retained legacy compatibility path has an owner, reason, and retirement condition.
• The repo no longer exposes obsolete record concepts in primary docs, agent context, or public APIs unless intentionally supported.
• make agent-pr-gate passes after cleanup.