Minor updates for worksplit_gpu with comfy-aimdo

[rattus128]

Upcoming comfy-aimdo has direct multi-GPU support (not yet released). This aimdo support symmetric device init so pass all devices to aimdo for upfront init (previous versions depended on a lazy init of non-primary GPUs). Fix the vbars_analyze call to be per-GPU and not report free-able VRAM as a global pool.

wheel that can be used here:

https://github.com/Comfy-Org/comfy-aimdo/actions/runs/24448365803

Example test conditions:

Linux, 2x4090
qwen 38GB (2512) 1328x1328 CFG=4 second run performance.

pytorch version: 2.11.0+cu130
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 4090 : cudaMallocAsync
Device: cuda:1 NVIDIA GeForce RTX 4090 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 115850.0
Using pytorch attention
aimdo: /project/src-posix/cuda-funchooks.c:126:DEBUG:aimdo_setup_hooks: hooks successfully installed
aimdo: /project/src/control.c:144:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 4090 (VRAM: 24078 MB)
aimdo: /project/src/control.c:144:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 4090 (VRAM: 24080 MB)
DynamicVRAM support detected and enabled
Python version: 3.12.3 (main, Mar  3 2026, 12:15:18) [GCC 13.3.0]
ComfyUI version: 0.18.1
comfy-aimdo version: 0.2.13.dev15
comfy-kitchen version: 0.2.8

1 GPU:

got prompt
Model QwenImage prepared for dynamic VRAM loading. 38968MB Staged. 0 patches attached.
100%|██████████████████████████████████████████████████████████████████████████████| 20/20 [01:02<00:00, 3.13s/it]
2 GPU:

Model QwenImage prepared for dynamic VRAM loading. 38968MB Staged. 0 patches attached.
Model QwenImage prepared for dynamic VRAM loading. 38968MB Staged. 0 patches attached.
100%|███████████████████████████████████████████████████████████████████████████████████| 20/20 [00:32<00:00, 1.62s/it]