Memory problem with MAC since ComfyUI update

### Expected Behavior

Previously, I never have memory problem with ComfyUI nor problem with disk.

### Actual Behavior

Now, especially when it comes to vae decoding, the memory increases rapidly and it takes up the whole disc as there was no more space on the disc. I am just doing a basic workflow with a batch size of 3 images 1024 * 1024. It crashed two times the computer. It is so rare on Mac. Now I get from time to time memory problem this was never the case in the past. There is more than 20G available on the disc. Sometimes the problem occurred when we start the workflow and it starts to load the model. We can no longer load two models in the same workflow without any problem. It looks like the disk and memory suddenly skyrocket.
It seems that comfyUI doesn't clean the memory and the disk or partially after each image render.
I have 16 Giga of RAM

### Steps to Reproduce

I can not send you step to reproduce it as it is a memory problem

### Debug Logs

```powershell
got prompt
model_type EPS
Using split attention in VAE
Using split attention in VAE
Requested to load SDXLClipModel
Loading 1 new model
Requested to load CLIPVisionModelProjection
Loading 1 new model
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:42<00:00, 10.60s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 400.22 seconds
got prompt
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:42<00:00, 10.71s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 99.16 seconds
got prompt
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:42<00:00, 10.72s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 76.92 seconds
got prompt
Prompt executed in 0.02 seconds
got prompt
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:41<00:00, 10.36s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 81.67 seconds
got prompt
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:42<00:00, 10.51s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 71.60 seconds
got prompt
Using split attention in VAE
Using split attention in VAE
Requested to load AutoencoderKL
Loading 1 new model
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:14<00:00, 3.58s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 38.31 seconds
got prompt
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:41<00:00, 10.50s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 94.84 seconds
got prompt
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:41<00:00, 10.47s/it]
Requested to load AutoencoderKL
Loading 1 new model
Prompt executed in 96.48 seconds
got prompt
Requested to load SDXL
Loading 1 new model
100%|█████████████████████████████████████████████| 4/4 [00:41<00:00, 10.49s/it]
Requested to load AutoencoderKL
Loading 1 new model
```

### Other

_No response_