When the consumer stops after yielding part of the messages due at one trigger time (e.g. messenger:consume --limit 1, SIGINT, fatal error), the next process rebuilt the heap with each trigger's run strictly after the checkpointed time, silently dropping the un-yielded messages at that time.

On the first heap build of a new instance with a non-warmup checkpoint, probe triggers one microsecond before $lastTime so they re-emit their entries due at $lastTime. The existing skip logic ($time == $lastTime && $index <= $lastIndex) then filters out already-yielded indices and lets the un-processed remainder through.

Reproduces with two RecurringMessage::every('30 seconds', …) consumed under --limit 1: "first" runs on every restart, "second" never runs without this patch.

TDD verification

• RED (new test without prod fix):

F 1/1 (100%)

1) Symfony\Component\Scheduler\Tests\Generator\MessageGeneratorTest::testCheckpointWithMultipleRecurringMessagesAtSameTriggerTime
Failed asserting that two arrays are identical.
--- Expected
+++ Actual
@@ @@
-Array &0 (
-    0 => stdClass Object &00000000000001be0000000000000000 (
-        'id' => 'second'
-    )
-)
+Array &0 ()

FAILURES! Tests: 1, Assertions: 3, Failures: 1.

• GREEN (new test with prod fix): OK (1 test, 3 assertions)
• Full MessageGeneratorTest: OK (18 tests, 81 assertions)
• Full Scheduler suite: OK (144 tests, 773 assertions) — no regression.