chore(lore-0167): activate; link the originating I-usd-rate-table note both ways

[karczuRF]

Activates 0167 (prices.usd_rate + peg-rate population) and repairs the lineage to the note the idea came from.

Why this jumps the queue

0167 is the only open task whose input data expires.

oracle_prices is pruned by monthly partition drop at INTERVAL 13 MONTH (cleanup-worker/src/lib.rs:24), and coverage starts ~2025-09 — so 202509 ages out around 2026-10/11. Every month this waits, a month of depeg-aware history is lost permanently and cannot be reconstructed.

Everything else on the board (0170, 0171, 0172, 0142) is static — those defects will be exactly as fixable next month.

And it can't be shortcut at the view layer: joining oracle_prices directly would make the published series mutate, with a bucket reading 0.9993 silently reverting to 1.0000 as rows age out. That's why views.sql:47 forbids the join. The rate must be snapshotted into a forever-retained table.

Downstream is ready

0165 shipped to prod today and is publishing 891 daily USDC buckets at a hardcoded $1 with method = 'peg'. Its three forward-compat requirements all landed — fallback semantics in the test rather than the literal 1, the method provenance column, the header naming 0168 by ID — so 0168 is a one-expression change against this table rather than a rewrite.

Worth restating the actual argument for it: not hypothetical depegs, but that two of our own surfaces already disagree. The oracle enrichment tier prices a TF/USDC candle at close × 0.9993 while price_usd_series publishes USDC = 1.0000 for the same bucket.

The lineage fix

The reasoning record lives in I-usd-rate-table.md under archived task 0144. It had spawns: [], and 0167 never cited it — so the note, including why the schema-wide refactor was rejected, was reachable only by knowing it existed in an archived task's notes directory. That is how a settled decision gets accidentally relitigated.

Now linked both ways: spawns: ["0167", "0168"] on the note, and the note in 0167's links.

Scope guard restated

The schema-wide refactor stays rejected per 0151. close_usd untouched, nothing published becomes NULL, and the key is natural identity, not quote_asset_id — 0139 is confirmed genuine collisions, measured today at 3,281 asset_ids across 6,568 identities, which a quote_asset_id-keyed table would have inherited wholesale.