fix(lore-0059): rollup timestamp-shadowing fix + full-chain version-propagation test

[karczuRF]

Task

lore-0059 — MV rollup-chain version propagation under enriched _1m re-inserts. Unblocked once 0051 landed the production MV DDL.

What this does

Closes 0059's remaining ACs by verifying the real shipped rollup chain against a live ClickHouse, and fixes a correctness bug the verification surfaced.

New: full-chain integration test

packages/prices-clickhouse/tests/rollup_chain_it.rs (#[ignore], runs against the pinned docker CH):

• Drives the real rollups.sql chain end-to-end _1m → _15m → _1h → _4h → _1d → _1w → _1M via SYSTEM REFRESH VIEW + poll.
• Proves an enrichment re-INSERT (version+1, volume_quote_usd filled) propagates to every grain with no under-count and no double-count (volume_base stays 30; volume_quote_usd 0 → 300; version 1 → 2 wins at every grain).
• Adds a preroll.sql full-range pass asserting the same OHLC correctness.

Bug fixed: argMin/argMax timestamp shadowing

The as-shipped rollups.sql and preroll.sql alias the bucket key toStartOfInterval(timestamp, …) AS timestamp, which shadows the source timestamp column. The bare timestamp inside argMin(open) / argMax(close) / argMax(close_usd) then resolved to the constant bucket-start, so open/close/close_usd tie-broke to an arbitrary row instead of the true first/last by time (volumes, high, low were unaffected). Fixed by qualifying the source column via FROM … AS t + t.timestamp.

• current.sql was already correct (AS c + c.timestamp).
• Schema-overview §3.2 reference DDL updated to match.

Review follow-ups (post-review)

Two issues raised in code review, both addressed in this branch:

• Pin local/CI ClickHouse to the EXACT prod version. The test was being validated against clickhouse-server:25.6 (25.6.13.41) while ch-prod-01 runs 26.3.10.60 — a 25.x → 26.x major gap, so "green locally" did not cover the engine the shipped DDL actually runs on. docker-compose.yml is now pinned to clickhouse/clickhouse-server:26.3.10.60; the full prices-clickhouse suite (rollup_chain, current_mv, views, lib) was re-verified green against it. Policy going forward: local/CI matches the live ch-prod-01 version — re-check SELECT version() on prod before bumping the pin. This complements, does not replace the live re-verify owed by 0071.
• Deterministic test anchor. insert_bucket re-evaluated toStartOfInterval(now(), …) per INSERT, so a wall-clock 15-minute boundary crossing between the un-enriched and enriched batches would anchor them to different buckets → broken ReplacingMergeTree dedup and a flaky single-bucket assertion. bucket_anchor() now freezes the boundary once as an integer epoch (toUInt64(toUnixTimestamp(…)) → toDateTime(<n>)), reused across every INSERT and robust across CH versions/timezones (avoids formatDateTime specifier portability).

⚠️ Live follow-up — 0071

The buggy DDL is already applied on ch-prod-01 (under 0051, 2026-06-22). Spawned lore-0071 to re-create the six live MVs from the corrected DDL + recompute mis-rolled buckets. Not done here (prepare-not-deploy).

Testing

docker compose up -d clickhouse                                       # CH 26.3.10.60 (matches ch-prod-01)
cargo test -p prices-clickhouse --test rollup_chain_it -- --ignored   # 2 passed
cargo test -p prices-clickhouse --test current_mv_it --test views_it -- --ignored   # 3 passed
cargo test -p prices-clickhouse --lib                                  # 5 passed (incl. statement-count guards)
cargo clippy -p prices-clickhouse --tests                             # clean

[karczuRF]

✅ Full suite verified against the prod ClickHouse version

Re-ran the prices-clickhouse suite against a local CH pinned to 26.3.10.60 — the exact version running on ch-prod-01 (confirmed SELECT version() → 26.3.10.60). Working tree was clean, so this is the committed branch state.

Check	Result
--lib unit tests (statement-count guards)	5 passed
rollup_chain_it (--ignored)	2 passed
current_mv_it (--ignored)	1 passed
views_it (--ignored)	2 passed
clippy --tests	clean
rollup_chain_it repeated 6× back-to-back	6/6 passed

Notes:

• No regression — every test green on 25.6 is green on 26.3.10.60.
• Determinism fix validated on prod version: the anchor SQL (toUInt64(toUnixTimestamp(toStartOfInterval(now(), INTERVAL 15 MINUTE))) frozen once → rebuilt as toDateTime(<epoch>)) round-trips correctly and yields 3 distinct per-minute rows (arrayUniq = 3) — the premise the ReplacingMergeTree dedup / single-bucket assertions rely on. The 6× repeat specifically stresses the old 15-minute-boundary flakiness.

Caveat: this validates the test + its SQL on the prod engine version, not the shipped rollup DDL applied to the live ch-prod-01 cluster — that live re-apply + verify remains 0071's job.