scope: on unified, make sure to unwatch all PIDs once they've been moved to the cgroup scope

Commit 428a9f6f1d0396b9eacde2b38d667cbe3f15eb55 freed u->pids which is
problematic since the references to this unit in m->watch_pids were no more
removed when the unit was freed.

This patch makes sure to clean all this refs up before freeing u->pids by
calling unit_unwatch_all_pids().

src/core/scope.c

@@ -375,10 +375,6 @@ static int scope_start(Unit *u) {
                 return r;
         }
 
-        /* Now u->pids have been moved into the scope cgroup, it's not needed
-         * anymore. */
-        u->pids = set_free(u->pids);
-
         s->result = SCOPE_SUCCESS;
 
         scope_set_state(s, SCOPE_RUNNING);
@@ -386,7 +382,13 @@ static int scope_start(Unit *u) {
         /* Set the maximum runtime timeout. */
         scope_arm_timer(s, usec_add(UNIT(s)->active_enter_timestamp.monotonic, s->runtime_max_usec));
 
-        /* Start watching the PIDs currently in the scope */
+        /* On unified we use proper notifications hence we can unwatch the PIDs
+         * we just attached to the scope. This can also be done on legacy as
+         * we're going to update the list of the processes we watch with the
+         * PIDs currently in the scope anyway. */
+        unit_unwatch_all_pids(u);
+
+        /* Start watching the PIDs currently in the scope (legacy hierarchy only) */
         (void) unit_enqueue_rewatch_pids(u);
         return 1;
 }

src/libsystemd/sd-event/sd-event.c

@@ -3725,7 +3725,7 @@ _public_ int sd_event_run(sd_event *e, uint64_t timeout) {
                 this_run = now(CLOCK_MONOTONIC);
 
                 l = u64log2(this_run - e->last_run);
-                assert(l < sizeof(e->delays));
+                assert(l < ELEMENTSOF(e->delays));
                 e->delays[l]++;
 
                 if (this_run - e->last_log >= 5*USEC_PER_SEC) {