Drop the query planner statistics tables

[epompeii]

What changed

• New migration 2026-09-14-120000_drop_planner_statistics drops sqlite_stat1 and sqlite_stat4 (DROP TABLE IF EXISTS, since a database that never ran ANALYZE has neither and a build without ENABLE_STAT4 never creates the second). Its down.sql is a comment: the statistics are not restored, because re-running a limited ANALYZE would reintroduce the problem. The 2026-08-17 migration is left as is; applied migrations are history.
• New test migrations_leave_no_planner_statistics_behind in bencher_schema asserts that no sqlite_stat% table exists after run_migrations.
• One rule in AGENTS.md, next to the other database rules: migrations never run ANALYZE or PRAGMA optimize, and nothing creates the sqlite_stat tables; the query planner runs without statistics.
• Changelog line under a new Pending v0.6.13 section.

Why

The 2026-08-17 migration ended with a limited ANALYZE. The sampled statistics overstate how many rows share a key on indexes whose leading rows cluster on one value. Trusting them, the planner scans the whole alert table for a status filtered alerts list and walks every head version for a perf query. A full ANALYZE does not help: it keeps the alert scan and adds scans on other query shapes. Without statistics the planner falls back on its structural heuristics, which pick the indexes, as it did before the statistics were introduced.

How it was tested

• Red then green: the new test was run on devel before the migration existed and failed with found ["sqlite_stat1", "sqlite_stat4"], since ANALYZE creates both tables even on an empty database. With the migration it passes.
• Round trip: lib/api_projects/tests/metric_migration.rs reverts and reapplies the migrations, so the comment-only down.sql and the re-drop after the re-run ANALYZE are both exercised.
• cargo nextest run --all-features --profile ci, cargo test --doc --all-features, cargo clippy --no-deps --all-targets --all-features -- -Dwarnings, cargo check --no-default-features, cargo fmt --all -- --check.

[github-actions]

Bencher Report

Project	Bencher
Branch	drop-planner-statistics
Testbed	intel-v1

Click to view all benchmark results

Benchmark	Latency	Benchmark Result microseconds (µs) (Result Δ%)	Upper Boundary microseconds (µs) (Limit %)
Adapter::Json	📈 view plot 🚷 view threshold	5.23 µs (+9.64%) Baseline: 4.77 µs	5.65 µs (92.67%)
Adapter::Magic (JSON)	📈 view plot 🚷 view threshold	5.03 µs (+8.66%) Baseline: 4.63 µs	5.40 µs (93.10%)
Adapter::Magic (Rust)	📈 view plot 🚷 view threshold	28.20 µs (+7.91%) Baseline: 26.13 µs	28.56 µs (98.75%)
Adapter::Rust	📈 view plot 🚷 view threshold	4.68 µs (+25.70%) Baseline: 3.72 µs	5.47 µs (85.52%)
Adapter::RustBench	📈 view plot 🚷 view threshold	4.68 µs (+25.82%) Baseline: 3.72 µs	5.46 µs (85.71%)

🐰 View full continuous benchmarking report in Bencher