fix(api_server): mark rejected API_SERVER_KEY as non-retryable fatal error

When _api_key_passes_startup_guard() rejects the key (missing,
placeholder/too short, or fail-closed unverifiable strength), connect()
returned a bare False with no fatal-error info. gateway.run's reconnect
watcher treats that as transient and re-queues with backoff forever —
each retry re-instantiating the adapter and its ResponseStore sqlite
connection. Observed in production (#37011): ~501 leaked connections
(1002 fds) over ~2.5 days until EMFILE made the whole gateway
unresponsive.

Set a non-retryable fatal error (api_server_key_invalid) in connect()
when the guard rejects, covering all three rejection branches, so the
platform drops from the reconnect queue; recover with
`/platform resume api_server` after fixing the key. Same treatment as
the port-conflict guard (api_server_port_in_use, #65665 / bda8bd76a8).

Tests mirror the port-conflict precedent: each rejection path asserts
connect() is False, has_fatal_error True, fatal_error_retryable False,
and fatal_error_code api_server_key_invalid, plus a strong-key control.

Re-implementation of #38803 by @cifangyiquan against current main —
their patch targeted the old inline guard in connect() which was since
extracted to _api_key_passes_startup_guard() (and gained the
fail-closed branch in 683059feb5), so the original diff no longer
applies. Their production diagnosis and non-retryable direction
preserved.

Refs: #38803, #37011

tests/gateway/test_api_server.py

@@ -4868,3 +4868,76 @@ class TestApiKeyStartupGuardFailsClosed:
     def test_missing_key_still_refused(self):
         """Control: the empty-key branch is unchanged."""
         assert self._guard("") is False
+
+
+class TestKeyRejectionSetsNonRetryableFatalError:
+    """Each startup-guard rejection must set a non-retryable fatal error so
+    the reconnect watcher drops the platform from the retry queue instead of
+    looping indefinitely.
+
+    Previously connect() returned bare ``False``, which gateway.run treated
+    as retryable — re-queueing every backoff interval forever and
+    re-instantiating the adapter (with its ResponseStore sqlite connection)
+    each retry (#38803: ~501 leaked connections / 1002 fds over 2.5 days,
+    ending in EMFILE for the whole gateway). Mirrors the port-conflict
+    precedent (test_port_conflict_sets_non_retryable_fatal_error, #65665).
+    """
+
+    @staticmethod
+    def _make_adapter(key, monkeypatch):
+        monkeypatch.delenv("API_SERVER_KEY", raising=False)
+        return APIServerAdapter(
+            PlatformConfig(
+                enabled=True,
+                extra={"host": "127.0.0.1", "port": 0, "key": key},
+            )
+        )
+
+    @staticmethod
+    async def _assert_key_rejection_is_fatal(adapter):
+        try:
+            assert await adapter.connect() is False
+            assert adapter.has_fatal_error is True
+            assert adapter.fatal_error_retryable is False
+            assert adapter.fatal_error_code == "api_server_key_invalid"
+            assert "API_SERVER_KEY" in (adapter.fatal_error_message or "")
+        finally:
+            await adapter.disconnect()
+
+    @pytest.mark.asyncio
+    async def test_missing_key_sets_non_retryable_fatal_error(self, monkeypatch):
+        adapter = self._make_adapter("", monkeypatch)
+        await self._assert_key_rejection_is_fatal(adapter)
+
+    @pytest.mark.asyncio
+    async def test_weak_key_sets_non_retryable_fatal_error(self, monkeypatch):
+        """Placeholder / <16-char keys are rejected by has_usable_secret."""
+        adapter = self._make_adapter("test", monkeypatch)
+        await self._assert_key_rejection_is_fatal(adapter)
+
+    @pytest.mark.asyncio
+    async def test_unverifiable_key_sets_non_retryable_fatal_error(self, monkeypatch):
+        """The fail-closed branch: a strong key whose strength cannot be
+        verified (hermes_cli.auth unimportable) must also be non-retryable —
+        the install won't repair itself between retries."""
+        adapter = self._make_adapter("a" * 40, monkeypatch)
+        real_import = __import__
+
+        def _blocked(name, *args, **kwargs):
+            if name == "hermes_cli.auth":
+                raise ImportError("simulated: hermes_cli.auth unavailable")
+            return real_import(name, *args, **kwargs)
+
+        with patch("builtins.__import__", _blocked):
+            await self._assert_key_rejection_is_fatal(adapter)
+
+    @pytest.mark.asyncio
+    async def test_strong_key_leaves_no_fatal_error(self, monkeypatch):
+        """Control: a successful connect() must not carry a fatal error."""
+        adapter = self._make_adapter("a" * 40, monkeypatch)
+        try:
+            assert await adapter.connect() is True
+            assert adapter.has_fatal_error is False
+            assert adapter.fatal_error_retryable is True
+        finally:
+            await adapter.disconnect()