feat: add scheduler and mail outbox

SMTP adapter with TLS and a fake, outbox with retries, backoff, rate limit and restart safety, scheduler tick with job-run records started by gunicorn, and a send_test_mail command.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N3t4WL7bd9bdHHNjEsxzqC

portal/tests/test_scheduler.py

@@ -0,0 +1,104 @@
+"""The scheduler: one tick runs the jobs and records each run (AD-12)."""
+
+import threading
+from datetime import timedelta
+
+import pytest
+from django.utils import timezone
+
+from portal.mailer import services as mailer
+from portal.scheduler import services
+from portal.scheduler.models import JobRun
+from portal.scheduler.runner import Runner
+
+pytestmark = pytest.mark.django_db(transaction=True)
+
+
+def test_a_tick_runs_the_mail_job_and_records_the_run(adapters):
+    mailer.enqueue("test", "lena@example.test")
+    services.tick()
+    assert len(adapters.mailer.sent) == 1
+    run = JobRun.objects.get(job="mail.flush")
+    assert run.status == JobRun.Status.OK and run.finished_at
+    assert run.detail == "sent=1 retried=0 failed=0"
+
+
+def test_an_idle_tick_is_recorded_too_and_repeating_it_is_harmless(adapters):
+    services.tick()
+    services.tick()
+    assert JobRun.objects.filter(job="mail.flush").count() == 2
+    assert adapters.mailer.sent == []
+
+
+def test_a_restart_sends_nothing_twice(adapters):
+    mailer.enqueue("test", "lena@example.test")
+    services.tick()
+    services.tick(timezone.now() + timedelta(minutes=5))  # "after a restart"
+    assert len(adapters.mailer.sent) == 1
+
+
+def test_a_failing_job_is_recorded_and_the_others_still_run():
+    ran = []
+
+    def broken(now):
+        raise RuntimeError("secret detail that must not be stored")
+
+    jobs = (
+        services.Job("first", lambda now: ran.append("first") or "fine"),
+        services.Job("broken", broken),
+        services.Job("last", lambda now: ran.append("last") or "also fine"),
+    )
+    services.tick(jobs=jobs)
+    assert ran == ["first", "last"]
+    broken_run = JobRun.objects.get(job="broken")
+    assert broken_run.status == JobRun.Status.ERROR and broken_run.detail == "RuntimeError"
+    assert JobRun.objects.get(job="last").status == JobRun.Status.OK
+
+
+def test_old_run_records_are_removed():
+    now = timezone.now()
+    JobRun.objects.create(job="x", started_at=now - timedelta(days=services.KEEP_RUNS_DAYS + 1))
+    JobRun.objects.create(job="x", started_at=now - timedelta(days=1))
+    services.tick(now, jobs=())
+    assert JobRun.objects.filter(job="x").count() == 1
+
+
+# -- the thread -----------------------------------------
+
+
+def test_the_thread_ticks_repeatedly_and_stops():
+    ticked = threading.Semaphore(0)
+    runner = Runner(tick=ticked.release, interval=0.01)
+    runner.start()
+    assert ticked.acquire(timeout=2) and ticked.acquire(timeout=2)
+    runner.stop()
+    assert not runner._thread.is_alive()
+
+
+def test_a_failing_tick_does_not_stop_the_thread():
+    calls = threading.Semaphore(0)
+
+    def tick():
+        calls.release()
+        raise RuntimeError("boom")
+
+    runner = Runner(tick=tick, interval=0.01)
+    runner.start()
+    assert calls.acquire(timeout=2) and calls.acquire(timeout=2)
+    runner.stop()
+
+
+def test_starting_twice_runs_only_one_thread():
+    runner = Runner(tick=lambda: None, interval=0.05)
+    runner.start()
+    first = runner._thread
+    runner.start()
+    assert runner._thread is first
+    runner.stop()
+    assert sum(1 for t in threading.enumerate() if t.name == "scheduler") == 0
+
+
+def test_the_default_interval_is_thirty_seconds():
+    from portal.scheduler.runner import TICK_SECONDS, runner
+
+    assert TICK_SECONDS == 30 and runner._interval == 30