fix(host_agent): an agent already offline stays offline through a Steward restart (#5719)

Counting every agent's silence from when Steward started watching again
(#5441) read an agent that was already gone as just seen: its agent-offline
alert resolved at 0 s and fired again three minutes later, two false emails
per restart (VDNT-Docker03, four times in a day). An agent whose last
recorded seconds_since_seen was already past the offline window now keeps
counting its whole silence; one that was fine when Steward stopped still
counts from the restart, so Steward's own downtime still fires nothing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

tests/plugins/host_agent/test_scheduler.py

@@ -71,3 +71,23 @@ def test_arrivals_are_kept_per_app():
     app, other = SimpleNamespace(), SimpleNamespace()
     arrivals(app)["h1"] = NOW
     assert arrivals(app) == {"h1": NOW} and arrivals(other) == {}
+
+
+def test_an_agent_already_offline_stays_offline_through_a_restart():
+    """#5719: VDNT-Docker03's agent was silent for hours when Steward
+    restarted, and counting it from the restart read it as just seen: its
+    alert resolved, then fired again three minutes later, two emails a
+    restart. An agent already past the offline window keeps its whole
+    silence; one that was fine when Steward stopped still counts from the
+    restart (#5441)."""
+    back = NOW - timedelta(minutes=1)
+    regs = [FakeReg("h1", NOW - timedelta(hours=5), NOW - timedelta(days=9)),
+            FakeReg("h2", NOW - timedelta(minutes=10), NOW - timedelta(days=9))]
+    assert seconds_since_seen(regs, NAMES, NOW, back, already_silent={"nas"}) == [
+        ("nas", 5 * 3600.0), ("web", 60.0)]
+
+
+def test_an_offline_agent_that_reports_after_the_restart_is_back():
+    back = NOW - timedelta(minutes=1)
+    regs = [FakeReg("h1", NOW - timedelta(seconds=15), NOW - timedelta(days=9))]
+    assert seconds_since_seen(regs, NAMES, NOW, back, already_silent={"nas"}) == [("nas", 15.0)]