fix(discord): watchdog de gateway — procesul viu pe un gateway mort nu mai trece neobservat

Puntea a tacut ~7 ore fara ca nimic sa semnaleze: discord.py memoreaza
`resume_gateway_url` primit la ultimul READY si il refoloseste la fiecare
reconectare. Cand gatewayul regional (gateway-us-east-1a) a inceput sa dea
503 la handshake, botul a reincercat la infinit acelasi host mort, cu backoff
pana la ~15 minute. Procesul era viu, deci nici systemd nici dashboardul nu
aveau ce vedea, iar `gateway.discord.gg` raspundea normal tot timpul.

GatewayWatchdog numara de cat timp e legatura jos (on_connect / on_resumed /
on_ready o ridica, on_disconnect o coboara, iar reincercarile esuate nu
reseteaza ceasul). Peste `GATEWAY_WATCHDOG_S` (implicit 300s) alerteaza, iese
cu codul 3 si lasa systemd sa reporneasca — restartul e singurul lucru care
forteaza un IDENTIFY nou pe gateway.discord.gg. Un prag <= 0 il dezactiveaza.

O iesire la 300s ramane sub StartLimitBurst=5/5min, deci bucla de restart nu
poate ajunge sa lase unitul `failed`. `_hard_exit_after` acopera cazul in care
`close()` se blocheaza pe socketul mort.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q4uzvgm7AyJch5WH8QHRhY

proxmox/lxc171-claude-agent/discord-bridge/tests/test_bot.py

@@ -421,3 +421,54 @@ async def test_cererea_de_aprobare_posteaza_butoane_in_fir(bridge, allowed):
     assert "rm -rf /tmp/x" in ch.texts[0]
     if bot.discord is not None:
         assert ch.sent[0].kwargs.get("view") is not None
+
+
+# ---------------------------------------------------------------- watchdog
+
+
+class FakeClock:
+    def __init__(self):
+        self.t = 0.0
+
+    def __call__(self):
+        return self.t
+
+
+def test_watchdogul_tace_cat_timp_gatewayul_e_sus():
+    clock = FakeClock()
+    wd = bot.GatewayWatchdog(300.0, now=clock)
+    wd.on_up()
+    clock.t = 10_000.0  # bot linistit ore intregi: niciun eveniment intre timp
+    assert wd.downtime() == 0.0
+    assert not wd.expired()
+
+
+def test_watchdogul_iarta_o_reconectare_scurta():
+    clock = FakeClock()
+    wd = bot.GatewayWatchdog(300.0, now=clock)
+    wd.on_down()
+    clock.t = 4.0
+    wd.on_up()  # RESUMED
+    clock.t = 900.0
+    assert not wd.expired()
+
+
+def test_watchdogul_expira_dupa_prag_iar_reincercarile_nu_reseteaza_ceasul():
+    clock = FakeClock()
+    wd = bot.GatewayWatchdog(300.0, now=clock)
+    wd.on_down()
+    for t in (60.0, 120.0, 240.0):  # fiecare handshake 503 mai da un on_disconnect
+        clock.t = t
+        wd.on_down()
+        assert not wd.expired()
+    clock.t = 300.0
+    assert wd.expired()
+    assert wd.downtime() == 300.0
+
+
+def test_watchdogul_se_poate_dezactiva():
+    clock = FakeClock()
+    wd = bot.GatewayWatchdog(0.0, now=clock)
+    wd.on_down()
+    clock.t = 100_000.0
+    assert not wd.expired()