Declare this machine's GPUs from serve, and refuse a bad limit as a flag

GPU count is not detected, so FLOW_GPUS was 0 on a fresh install and a node
asking for one was silently clamped to zero and ran concurrently with every
other. Setting the variable serialised them, but it was an environment
variable only — `serve` had --max-runs and --max-workers and no --gpus.
The clamp warning now names the flag when nothing here declares a card.

The same flags are written into the environment before the settings are
built, so a value they refused died in a pydantic import naming no flag.
They are checked where they are typed instead.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019Hra4ndWMCLU5F3KjUuVAc

backend/tests/test_cli.py

@@ -520,6 +520,24 @@ def test_two_files_of_one_name_are_refused(tmp_path) -> None:
         _import(*_module_of(second), second)
 
 
+def test_a_serve_limit_is_refused_as_a_flag_not_as_a_traceback(capsys) -> None:
+    """These are written into the environment before the settings are built."""
+    import pytest
+
+    from fluksio.cli import _parser
+
+    parser = _parser()
+    assert parser.parse_args(["serve", "--max-workers", "2"]).max_workers == 2
+    # A machine may genuinely have no card, so zero is a number here.
+    assert parser.parse_args(["serve", "--gpus", "0"]).gpus == 0
+    assert parser.parse_args(["serve"]).gpus is None
+
+    for flag, value in (("--max-workers", "0"), ("--gpus", "-1")):
+        with pytest.raises(SystemExit):
+            parser.parse_args(["serve", flag, value])
+    assert "at least" in capsys.readouterr().err
+
+
 def test_run_and_sweep_take_what_to_sync() -> None:
     from fluksio.cli import _parser