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/fluksio/cli.py

@@ -21,6 +21,7 @@ import os
 import secrets
 import socket
 import sys
+from collections.abc import Callable
 from pathlib import Path
 from typing import Any
 
@@ -228,9 +229,26 @@ CONCURRENCY_FLAGS = {
     "max_workers": "FLOW_MAX_WORKERS",
     "max_cascades": "FLOW_MAX_CASCADES",
     "max_runs": "FLOW_MAX_RUNS",
+    "gpus": "FLOW_GPUS",
 }
 
 
+def _at_least(minimum: int) -> Callable[[str], int]:
+    """A flag's value, checked here rather than by the settings.
+
+    These are written into the environment before the settings are built, so a
+    number they refuse dies inside a pydantic import with no flag named in it.
+    """
+
+    def parse(text: str) -> int:
+        value = int(text)
+        if value < minimum:
+            raise argparse.ArgumentTypeError(f"is {value}, needs at least {minimum}")
+        return value
+
+    return parse
+
+
 #: What `serve` listens on when nobody says. Taken often enough — another
 #: engine, another framework's dev server — that dying on it is the first
 #: thing a zero-config start would hit.
@@ -426,25 +444,32 @@ def _parser() -> argparse.ArgumentParser:
     )
     serve.add_argument(
         "--max-runs",
-        type=int,
+        type=_at_least(1),
         default=None,
         metavar="N",
         help="batch runs driven at once (default 4, FLOW_MAX_RUNS)",
     )
     serve.add_argument(
         "--max-cascades",
-        type=int,
+        type=_at_least(1),
         default=None,
         metavar="N",
         help="cascades in flight at once (default 4, FLOW_MAX_CASCADES)",
     )
     serve.add_argument(
         "--max-workers",
-        type=int,
+        type=_at_least(1),
         default=None,
         metavar="N",
         help="python worker processes (default 4, FLOW_MAX_WORKERS)",
     )
+    serve.add_argument(
+        "--gpus",
+        type=_at_least(0),
+        default=None,
+        metavar="N",
+        help="GPUs on this machine a node may be given (default 0, FLOW_GPUS)",
+    )
     serve.set_defaults(func=cmd_serve)
 
     enroll = subparsers.add_parser(