[codex] bound DB normal nightly soak

[feichai0017]

Summary

• bound the nightly DB normal soak job to a hosted-runner-sized DB regression
• keep multi-tree atomic/view/reopen coverage while leaving long durability coverage to DB crash soak
• document why this job should stay smaller than the crash soak

Root Cause

The nightly DB normal soak job used 500k keys / 1M ops / 8 threads and then performed a full oracle verification across every named tree after reopen. On GitHub hosted runners this job has either timed out around the 45 minute job limit or been canceled by the runner shortly after start. The logs did not show a Holt panic or oracle mismatch.

Validation

• cargo run --manifest-path tools/soak/Cargo.toml --locked -- --mode db-normal --dir target/holt-soak-db-nightly-local --reset --duration-secs 600 --keys 20000 --ops 50000 --threads 4 --buffer-pool 64 --wal-sync false
• cargo test --test tree_smoke db_ --all-features --locked
• cargo test --test wal_tree_integration db_ --all-features --locked
• git diff --check