Continue the memory-layers lab, then build its proven designs on the lean ledger

The memory exclusion is lifted. We continue Codex’s unfinished memory-layers experiment (~/bridlewise-workspaces/memory-layers-v1) with agents, under that lab’s own AGENTS.md rules (we are its coordinator; only lab-assigned agents write there, UR-10), and THEN implement its proven designs in the harness ON TOP OF the new lean ledger (lean-v1, the per-project shards, the archive tier). Only findings and proven designs transfer, never lab code; harness writers do not edit memory code, schema, storage, recall or injection until a design is admitted and scheduled (the MEM tasks), and a defect found meanwhile is recorded with its MEM owner (U-M6). The transfer replaces, never wraps (UR-01); memory recall stays a function the prompt cache never sacrifices (UR-03). The user tier of U6 moves to MEM-8; its placement is U-M1.