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

[henry40408]

Summary

Every dashboard tick scanned 30 days of idx_query_logs_ts_metrics for its summary, and the top domain, client and upstream lists scanned their own indexes over 24 hours. Under the default retention those windows are the whole table, so a tick cost an index's length every 10 seconds.

This moves those readings onto the rollups from #280.

• Readers moved: summary_multi_since, timeline_since, traffic_lists_since, domain_stats_since and top_upstreams_since.
• How they read: each takes every whole unit (quarter hour or hour) from a rollup, and reads query_logs only for the part of a unit its window starts inside, through idx_query_logs_timestamp, in one statement. Nothing is read at the far end: the rollups are written in the same transaction as the rows, so the unit still filling up is complete.
• Summary windows: summary_multi_since tells its three windows apart by which arm a table row came from. A row belongs to the one window whose partial quarter it fills; the wider windows count that quarter through the rollup.
• Fine-grained timeline: timeline_since keeps counting the table when its bucket is finer than a quarter hour, which only happens while the log is a few hours old.
• Behaviour change: top upstreams now break ties by name, as the domain and client lists already did. Before this, ties came back in arbitrary order.
• Also affects Statistics: traffic_lists_since is shared with the Statistics page, so its two lists move too. stats_scan_since and the /api/stats/* metrics statements are the next PR.

Page misses

All three benches on the same database copies and BENCH_NOW=1789212354, before (db67506) and after. Every reading that changed is listed; the rest are identical.

370 677 rows
  dashboard  summary                                              2157       28
  dashboard  top domains + clients                                 748      115
  dashboard  top upstreams                                         212       37
  dashboard  FIRST RESPONSE TOTAL                                 3117      180
  dashboard  timeline (tick only)                                  293       16
  dashboard  TICK TOTAL (compute_snapshot)                        3408      190
  logs       domain suggestions                                   3973      557
  logs       SUM OF ALL READINGS                                103445   100029
  stats      range_stats (all readings, charts, both lists)       7923     2704
  stats      top_domains alone (domain index)                     3973      540
  stats      both lists alone                                     5770      551
  stats      PAGE TOTAL (whole visit)                             7937     2718

1 482 708 rows
  dashboard  summary                                              9480       61
  dashboard  top domains + clients                                 752      116
  dashboard  top upstreams                                         212       38
  dashboard  FIRST RESPONSE TOTAL                                10444      215
  dashboard  timeline (tick only)                                  332       16
  dashboard  TICK TOTAL (compute_snapshot)                       10774      227
  logs       domain suggestions                                   6192      559
  logs       SUM OF ALL READINGS                                374596   368963
  stats      range_stats (all readings, charts, both lists)      31870    11628
  stats      top_domains alone (domain index)                    16084     2111
  stats      both lists alone                                    22394     2152
  stats      PAGE TOTAL (whole visit)                            31883    11641

No reading in any of the three benches got worse. idx_query_logs_ts_domain_client and idx_query_logs_ts_upstream now serve nothing; they are dropped in the index-consolidation PR, once the Statistics scan has moved too.

Tests

• dashboard_readings_equal_a_recount_of_the_table (tests/stats_db_test.rs): three days of traffic starting 1 234 ms past an hour. Every reading is compared with the statement it replaced, run on the same rows, for windows starting before the data, on an hour, on a quarter, inside a quarter, inside an hour, and after the data. The timeline uses 60/600/1 800/3 600 s buckets, the lists use limits 5 and 1 000. On the old code, every comparison it reached matched until the top-upstream tie order, which the old statement left arbitrary.
• the_dashboard_readings_fold_rollups_rather_than_the_table (tests/stats_page_miss_test.rs): replaces three tests tied to the old indexes (the_dashboard_summary_is_one_metrics_scan, the_top_upstreams_never_read_the_log_table, a_short_window_reads_a_short_stretch_of_the_traffic_lists). Each reading must cost under a tenth of a scan over every row. Before the implementation it failed with summary read 107 pages where a scan over every row reads 168.
• Removed: the_traffic_lists_are_covered_by_their_index (src/db.rs), which asserted the plan of a statement that no longer exists.

Test plan

• cargo fmt --check
• cargo clippy --all-targets -- -D warnings
• cargo nextest run: 681 passed, 8 skipped
• cd e2e && cargo test --test specs: all spec cases passed
• cd e2e && cargo test --test e2e: 17 scenarios (17 passed), 85 steps (85 passed)
• All three page-miss benches before/after on both databases

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 98.38710% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 91.32%. Comparing base (db67506) to head (ab0981d).

Files with missing lines	Patch %	Lines
src/db.rs	98.38%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #281      +/-   ##
==========================================
+ Coverage   91.30%   91.32%   +0.02%     
==========================================
  Files          31       31              
  Lines       11587    11625      +38     
==========================================
+ Hits        10579    10617      +38     
  Misses       1008     1008              

☔ 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.