feat: add 1h and 1d lowest available price candles

[us-oyster]

Summary

Adds end-to-end 1-hour and 1-day OHLC price candles for canonical products. PriceAI now records the lowest eligible purchasable price every 15 minutes, serves cached public chart APIs, renders compact candles on /channels, and renders an interactive detail chart on /products/[id].

Price definition

• Lowest CNY offer that is visible, positive-priced, fresh, available, in stock or low stock, unexpired, and has a purchase URL.
• Includes only offers with no bulk restriction or min_order_quantity <= 1.
• Requires two consecutive valid confirmations at the same canonical product and price.
• Excludes shared_access, web_only_account, and domestic_mirror_site.
• Uses deterministic updated-time and offer-ID tie-breaking.
• This is an observed public asking price, not an actual transaction price.

Data model

• Adds idempotent 15-minute product_price_samples.
• Adds product_price_candles for 1h and Asia/Shanghai calendar-day 1d OHLC.
• Extends raw-offer confirmations with canonical product and valid-confirmation streak tracking.
• Reaggregates affected candles from samples, so delayed sampling preserves correct open/close ordering.
• Keeps samples for 180 days, hourly candles for 365 days, and daily candles without automatic expiry.
• Retention defaults to dry-run and only removes samples whose hourly and daily aggregates exist.
• Tables, internal view, and maintenance RPCs remain service_role only.

UI

• /channels: one global 1小时 / 1天 switch and one batch request for all matching products, with lightweight SVG candles on desktop and mobile.
• /products/[id]: dynamically imports lightweight-charts@5.2.0 after mount and supports crosshair, touch/mouse interaction, zoom, drag, reset, and OHLC/sample tooltips.
• Covers loading, retry, no-history, accumulating-data, and current-price-missing states.
• Uses red for up, green for down, gray for flat, plus explicit signed text and a fixed non-transaction disclaimer.

Performance

• Summary reads use two batched RPCs for all filtered products; there is no database or HTTP N+1.
• Both public APIs use the existing Cloudflare edge cache policy with a 300-second TTL and complete query-parameter cache keys.
• The channels view uses custom SVG and does not load the charting library.
• The full chart library is client-only and dynamically imported on product detail pages.

Validation

• npm run test:price-history
• npm run test:price-history-db (Docker PostgreSQL with the real schema.sql)
• npm run test:price-history-api
• npm run lint (one pre-existing warning)
• npm run typecheck
• npm test
• npm run check:performance
• npm run check:recovery-baseline
• node scripts/verify-supabase-recovery-baseline.mjs --docker
• npm run audit:licenses
• npm run build
• npm run build:cloudflare

Database coverage includes eligibility filters, two confirmations, multi/single-sample OHLC, delayed and duplicate sampling, product/currency isolation, gaps, Asia/Shanghai day boundaries, historical reads without a current quote, and protected retention.

Screenshots

Validated with deterministic local fixtures. No production data is included.

Channels desktop - 1-day candles

Channels mobile - 1-hour candles

Product detail desktop - 1-day chart and tooltip

Product detail mobile - 1-hour chart and tooltip

Additional states validated locally: no-history and current-price-missing.

Limitations

• CNY only.
• No historical backfill; candles begin accumulating after deployment.
• Supports only 1h and 1d.
• No transaction volume, executed-price claim, or prediction.
• Does not bypass login walls, CAPTCHA, WAF, or other collection restrictions.

Rollout

1. Apply 20260720220000_product_price_history.sql.
2. Deploy the application.
3. Manually trigger one action=sample run and verify samples/candles/API output.
4. Enable the 15-minute sampling workflow and daily retention step.

Rollback should stop the scheduled jobs first and roll back the application. Keep the new history tables in place; no destructive down migration is required.