Optimize subtree delete lookup

Replace the subtree LIKE predicate with exact-root and bounded child range deletes so SQLite can use the (scan_id, path) primary key instead of scanning every row in the scan.

Benchmarks on the real 7.07M-entry / cache showed old LIKE predicates taking ~17-23s regardless of subtree size, while the new range predicate scaled with subtree size: /Applications ~18.0s -> ~0.85s, /System ~17.7s -> ~0.73s, /Library ~20.7s -> ~0.27s, /opt ~21.4s -> ~0.07s, /private ~18.9s -> ~0.02s, /bin ~22.2s -> ~0.02s. /Users, which covers most of the scan, improved ~20.8s -> ~16.0s.

A synthetic actual DELETE benchmark with 1,001,001 rows deleting a 1,001-row subtree improved from ~114ms to ~19ms.

Author: orestesgaolin

disk_analyzer_cli/lib/src/storage/database.dart

@@ -307,10 +307,26 @@ class DiskDatabase {
   /// Delete all entries under a subtree path within a scan (inclusive).
   /// Uses exact path match + path prefix with '/' boundary to avoid prefix collisions.
   void deleteSubtree(int scanId, String subtreePath) {
-    _db.execute(
-      'DELETE FROM entries WHERE scan_id = ? AND (path = ? OR path LIKE ? || \'/%\')',
-      [scanId, subtreePath, subtreePath],
-    );
+    final childPrefix = subtreePath == '/' ? '/' : '$subtreePath/';
+    // '/' sorts immediately before '0', giving a tight upper bound for children.
+    final childUpperBound = subtreePath == '/' ? '0' : '${subtreePath}0';
+
+    _db.execute('BEGIN TRANSACTION');
+    try {
+      _db.execute('DELETE FROM entries WHERE scan_id = ? AND path = ?', [
+        scanId,
+        subtreePath,
+      ]);
+      _db.execute(
+        'DELETE FROM entries '
+        'WHERE scan_id = ? AND path >= ? AND path < ?',
+        [scanId, childPrefix, childUpperBound],
+      );
+      _db.execute('COMMIT');
+    } catch (e) {
+      _db.execute('ROLLBACK');
+      rethrow;
+    }
   }
 
   /// Get the depth of an entry in a scan.