feat(agentos): Implement Continuous "Tech Debt Radar" via Frontier RAG Sweeps

Description

Our reflection gates (ticket-intake and pr-review) are already highly optimized and correctly mandated to execute Semantic RAG capabilities (ask_knowledge_base / Memory Core) natively over grep.

However, the failure mode is that these gates are strictly reactive. They only perform validation sweeps relative to the localized context of whatever active ticket or PR they were assigned. This makes them inherently blind to widespread, ambient architectural debt (like orphaned test directories, abandoned migration scripts, or systemic config regressions) that lie completely outside their immediate execution scope.

The "Debt Radar" Workflow

This issue tracks the implementation of an autonomous, proactive architectural review workflow native to the Agent OS.

1. Active Repository Traversing: The background daemon operates without the constraints of localized ticket context. It proactively sweeps closed Epics and historical items inside resources/content/issues/ and resources/content/issue-archive/ against the active filesystem structure.
2. Memory Core Session Mining (Crucial): Beyond just reading markdown tickets, the daemon must heavily query previous Memory Core agent sessions. Scanning past agent thoughts and iterations provides the rich internal monologue detailing why debt accrued and mapping abandoned architectural loops.
3. Proactive Maintenance: Utilizing this deep context, the sweep routinely catalogs depreciated logic clusters and generates granular, localized cleanup tickets for the swarm backlog.
4. Frontier Model Execution: This meta-analysis must be executed strictly by Frontier Models (Gemini 3.1 Pro / Claude Opus 4.6), isolating massive architectural discovery scans from tactical SML capacities.

Origin Session ID: fcd5f5f2-34ab-4c1a-acb1-533b88e123b3