fix(fusion_clock): code-review hardening [19.0.5.0.1]

- _cron_generate: per-rule savepoint isolation (one bad rule can't abort the
  whole daily batch)
- fclk_attach_recurrence: clear an existing recurrence first (no orphaned rule
  generating forever)
- fclk_apply_planner_cell: collapse split rows (search was limit=1 after the
  UNIQUE drop, orphaning extras)
- fclk_release_shift: reject non-posted/open shifts (raw-POST guard)
- delete_open_shift: report success=false when nothing was deleted + JS surfaces it
- _generate: log before removing an empty recurrence
Tests added for collapse, re-attach, draft-release.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

fusion_clock/tests/test_recurrence.py

@@ -83,6 +83,21 @@ class TestRecurrence(TransactionCase):
         self.assertNotIn(date(2026, 6, 8), dates, "Leave day should be skipped")
         self.assertIn(date(2026, 6, 15), dates)
 
+    def test_reattach_recurrence_replaces_old_rule(self):
+        seed = self._seed(date(2026, 6, 1))
+        rule1 = self.Schedule.fclk_attach_recurrence(seed, {
+            'repeat_interval': 1, 'repeat_unit': 'week',
+            'repeat_type': 'x_times', 'repeat_number': 3})
+        rule1_id = rule1.id
+        rule2 = self.Schedule.fclk_attach_recurrence(seed, {
+            'repeat_interval': 1, 'repeat_unit': 'week',
+            'repeat_type': 'x_times', 'repeat_number': 2})
+        # The old rule must be gone (not left generating forever) and the seed
+        # must point at the new rule.
+        self.assertFalse(
+            self.env['fusion.clock.schedule.recurrence'].browse(rule1_id).exists())
+        self.assertEqual(seed.recurrence_id, rule2)
+
     def test_clear_recurrence_unlinks_rule_when_empty(self):
         seed = self._seed(date(2026, 6, 1))
         rule = self.Schedule.fclk_attach_recurrence(seed, {