perf(stats): fold the Statistics page's readings out of the statistics rollups

[henry40408]

Fifth of the rollup series (#278 benches, #279 log counter, #280 rollups, #281 dashboard).

What changes

• stats_scan_since — the Statistics page's one read for its breakdowns, latency and chart series — is now a single four-arm statement: query_stats_quarter from the earlier window's first whole quarter, query_stats_metrics_hour from the range's first whole hour, and the query_logs rows each window starts inside (via idx_query_logs_timestamp). A leading column tells the arms apart, so each window keeps its edge to the millisecond.
• API-only readers fold the same tables: window_metrics_since (breakdowns, highlights' latency), timeline_multi_since and hourly_heatmap_since (quarter rollup when bucket and offset are whole quarters; a table count otherwise).
• /api/stats/v2/timeline and /api/stats/v2/heatmap round tz_offset to the nearest 15 minutes. Every zone in use already is; the rounding keeps every API call on the rollup path.
• No statement names idx_query_logs_ts_metrics any more, which unblocks dropping it (next PR). Its plan and migration tests stay until then.

Page misses

stats_page_miss_bench, BENCH_NOW=1789212354, main (d801b9c) → this branch.

Reading	1.48 M rows (31d)	370 k rows (8d)
Statistics visit, 30d range	11 641 → 3 644	2 718 → 946
Statistics visit, 7d range	10 060 → 1 015	2 735 → 980
API timeline, 30d	9 476 → 45	2 153 → 13
API heatmap	6 030 → 38	1 387 → 13
API breakdowns, 30d	9 476 → 1 446	2 153 → 372

dashboard_page_miss_bench and logs_page_miss_bench are identical to main on both databases (tick 227 / 190, logs sum 368 963 / 100 029).

Tests

• statistics_readings_equal_a_recount_of_the_table — scan, window metrics, timeline (5 buckets) and heatmap against a direct count of query_logs, for window starts before the data, on an hour, on a quarter, inside each and after; heatmap windows before/equal/after the range; offsets 0, +8h, −5h, +5:30, +5:45, −9:30 and an unaligned 7 min. Breaking each table arm in turn (scan, window, timeline, heatmap) makes it fail.
• the_statistics_readings_fold_rollups_rather_than_the_table — fails on main (stats scan 107 pages vs 168 for a full scan).
• stats_v2_tz_offset_is_rounded_to_a_quarter_hour — fails with the rounding removed.
• Removed the_page_and_its_charts_are_one_metrics_scan and the_heatmap_reads_the_narrowest_index_that_covers_it, which asserted index choices no reader makes now.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 96.85039% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.36%. Comparing base (d801b9c) to head (efee13b).

Files with missing lines	Patch %	Lines
src/db.rs	96.80%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #282      +/-   ##
==========================================
+ Coverage   91.32%   91.36%   +0.03%     
==========================================
  Files          31       31              
  Lines       11625    11691      +66     
==========================================
+ Hits        10617    10681      +64     
- Misses       1008     1010       +2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.