Split the worker into a distribution of its own

A cluster or GPU host installs `pip install fluksio-worker` and gets the
agent and the runner, not psycopg, numpy and the MCP SDK. The engine
depends on it as a workspace member, so the file it launches node code
with is the same file a remote worker runs — which is what keeps a node
unable to tell the difference. Copying the two files by hand still works.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

backend/fluksio/api/routes/workers.py

@@ -15,12 +15,12 @@ from typing import Any
 
 from fastapi import APIRouter, Depends, HTTPException, Request, WebSocket
 from fastapi.responses import PlainTextResponse
+from fluksio_worker import worker_main
 from jwt.exceptions import InvalidTokenError
 from pydantic import BaseModel, Field
 
 from fluksio.api.deps import get_current_active_superuser, get_current_user
 from fluksio.core import security
-from fluksio.flow import worker_main
 from fluksio.flow.remote import PROTOCOL, RemoteWorker, RemoteWorkerHub
 
 logger = logging.getLogger(__name__)
@@ -103,7 +103,7 @@ def read_runtime() -> str:
     """The worker's own code, so a fresh host installs by fetching one file.
 
     It is the same module the engine's local workers run — deliberately
-    standard library only, and with nothing of the app importable in it.
+    standard library only, and with nothing of the engine importable in it.
     """
     return worker_main.__file__ and open(worker_main.__file__).read()