test: close coverage gaps + opt-in integration smoke

Unit-Tests (kein Docker nötig), 130 Tests gesamt:
- config: ${ENV}- und ~-Expansion in hf_home
- prompt_io: _pin_dns (DNS-Rebinding wird abgewiesen, andere Hosts unberührt,
  Resolver wird wiederhergestellt)
- actions: do_check (running/healthy, missing), do_start non-dry-run
  (Happy-Path + Readiness-Fehler mit --logs), _container_lock Force-Bypass
- cli: --max-tokens <= 0 abgelehnt, --expose/--no-expose

scripts/smoke.sh: opt-in End-to-End-Test gegen echten Docker + llama.cpp-Server
(--api-key-Round-Trip inkl. 401/200, start/check/chat/stream/stop, eigener
Container/Port, Cleanup-Trap).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

tests/test_cli.py

@@ -39,6 +39,22 @@ def test_chat_requires_prompt_source():
     parse_error(["--chat"])
 
 
+def test_max_tokens_nonpositive_rejected():
+    parse_error(["--start", "--max-tokens", "0"])
+    parse_error(["--start", "--max-tokens", "-5"])
+
+
+def test_max_tokens_positive_ok():
+    args = parse(["--start", "--max-tokens", "4096"])
+    assert args.max_tokens == 4096
+
+
+def test_expose_and_no_expose_flags():
+    assert parse(["--start", "--expose"]).expose is True
+    assert parse(["--start", "--no-expose"]).expose is False
+    assert parse(["--start"]).expose is None  # unset => config decides
+
+
 def test_chat_with_prompt_is_valid():
     args = parse(["--chat", "-p", "hello"])
     assert args.chat is True