Bound SQLite integrity checks for diagnostic attachments (#40688)

## Why

Diagnostic uploads skip the doctor attachment if report generation exceeds its
overall timeout. SQLite integrity scans can be delayed by large databases or
lock contention.

## What changed

- Give each database integrity scan a one-second deadline when collecting a
  diagnostic attachment.
- Interrupt scans and lock waits at the deadline, preserve any corruption rows
  already returned, and report an incomplete clean scan as a warning.
- Keep direct `codex doctor` integrity checks unbounded.

## Testing

- Cover interrupted scans, lock waits, retries without a deadline, and
  corruption results returned before interruption.

GitOrigin-RevId: 4817fb459ac53953669a9ba5f03397a6fc82c72e

codex-rs/state/src/sqlite.rs

@@ -291,12 +291,19 @@ impl SqliteConfig {
     }
 
     /// Open an existing Codex SQLite database without creating or modifying it.
-    pub async fn open_read_only_pool(&self, path: &Path) -> Result<SqlitePool, Error> {
-        let options = SqliteConnectOptions::new()
+    pub async fn open_read_only_pool(
+        &self,
+        path: &Path,
+        busy_timeout: Option<Duration>,
+    ) -> Result<SqlitePool, Error> {
+        let mut options = SqliteConnectOptions::new()
             .filename(path)
             .create_if_missing(false)
             .read_only(true)
             .log_statements(LevelFilter::Off);
+        if let Some(busy_timeout) = busy_timeout {
+            options = options.busy_timeout(busy_timeout);
+        }
         SqlitePoolOptions::new()
             .max_connections(1)
             .connect_with(options)