Fix ComfyUI prompt autoswitch

[thinmintdev]

Summary

• remove the env gate from ComfyUI switchover/pin routes
• change the mounted ComfyUI custom node to request image mode before forwarding /prompt instead of 403-blocking inference mode
• sync shipped ComfyUI custom nodes during install/deploy so CT 105 picks up node changes

Verification

• PYTHONPATH=src .venv/bin/python -m pytest tests/install/test_comfyui_scripts_shipped.py tests/comfyui tests/api/test_comfyui_proxy.py tests/api/test_comfyui_phase4.py tests/api/test_comfyui_fetch_route.py tests/api/test_features_health.py tests/api/test_v1_images.py tests/providers/test_comfyui.py tests/providers/test_comfyui_workflows.py tests/slots/test_gpu_arbiter.py tests/dispatcher/test_arbiter_dispatch.py -q
• cd ui && npm run typecheck
• cd ui && npm run build

Deploy

• deployed branch to CT 105 at 8e90a2e
• synced /mnt/ai-models/comfyui/custom_nodes/hal0_gpu_gate.py
• restarted hal0-slot@img.service and confirmed ComfyUI status reachable + custom node imported