[Bug]: ControlNet inpaint incorrectly cropped mask

### Checklist

- [X] The issue exists after disabling all extensions
- [X] The issue exists on a clean installation of webui
- [ ] The issue is caused by an extension, but I believe it is caused by a bug in the webui
- [X] The issue exists in the current version of the webui
- [X] The issue has not been reported before recently
- [ ] The issue has been reported before but has not been fixed yet

### What happened?

When using ControlNet inpainting with resize mode set to 'crop and resize' the black and white mask image passed to ControlNet is cropped incorrectly.

### Example:
Original image:

Inpaint settings, resolution is 1024x1024:

Cropped outputs stacked on top, mask is clearly misaligned and cropped:

### Steps to reproduce the problem

1. Go to img2img inpaint
2. Input the example image above
3. Paint mask over strings and yellow balloon
4. Set inpaint settings, resolution 1024*1024:
5. Set ControlNet settings:
6. Generate

### What should have happened?

The yellow balloon should have been painted out

Example when settings inpaint area to 'Whole picture':

### What browsers do you use to access the UI ?

Mozilla Firefox

### Sysinfo

[sysinfo-2024-02-21-20-13.json](https://github.com/lllyasviel/stable-diffusion-webui-forge/files/14364558/sysinfo-2024-02-21-20-13.json)

### Console logs

```Shell
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: f0.0.14v1.8.0rc-latest-184-g43c9e3b5
Commit hash: 43c9e3b5ce1642073c7a9684e36b45489eeb4a49
Launching Web UI with arguments:
Total VRAM 12287 MB, total RAM 32717 MB
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 3060 : native
VAE dtype: torch.bfloat16
Using pytorch cross attention
ControlNet preprocessor location: G:\AI\clean forge\webui\models\ControlNetPreprocessor
Loading weights [821aa5537f] from G:\AI\clean forge\webui\models\Stable-diffusion\autismmixSDXL_autismmixPony.safetensors
2024-02-21 20:30:57,792 - ControlNet - INFO - ControlNet UI callback registered.
Running on local URL: http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
model_type EPS
UNet ADM Dimension 2816
Startup time: 30.1s (prepare environment: 8.0s, import torch: 7.5s, import gradio: 2.8s, setup paths: 2.7s, initialize shared: 0.2s, other imports: 1.8s, load scripts: 3.5s, create ui: 1.1s, gradio launch: 2.7s).
Using pytorch attention in VAE
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
Using pytorch attention in VAE
extra {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_g.transformer.text_model.embeddings.position_ids', 'cond_stage_model.clip_l.logit_scale'}
Loading VAE weights specified in settings: G:\AI\clean forge\webui\models\VAE\sdxl_vae.safetensors
To load target model SDXLClipModel
Begin to load 1 model
Moving model(s) has taken 0.62 seconds
Model loaded in 11.8s (load weights from disk: 0.8s, forge solving config: 0.2s, forge instantiate config: 2.3s, forge load real models: 7.0s, load VAE: 0.4s, calculate empty prompt: 1.0s).
2024-02-21 20:32:41,665 - ControlNet - INFO - ControlNet Input Mode: InputMode.SIMPLE
2024-02-21 20:32:41,818 - ControlNet - INFO - Crop input image based on A1111 mask.
2024-02-21 20:32:42,022 - ControlNet - INFO - Crop input image based on A1111 mask.
2024-02-21 20:32:42,308 - ControlNet - INFO - Using preprocessor: inpaint_only
2024-02-21 20:32:42,309 - ControlNet - INFO - preprocessor resolution = 0.5
Warning torch.load doesn't support weights_only on this pytorch version, loading unsafely.
2024-02-21 20:32:43,469 - ControlNet - INFO - Current ControlNet FooocusInpaintPatcher: G:\AI\clean forge\webui\models\ControlNet\inpaint_v26.fooocus.patch
To load target model AutoencoderKL
Begin to load 1 model
Moving model(s) has taken 0.33 seconds
[Fooocus Patch Loader] 960 keys loaded, 0 remaining keys not found in model.
2024-02-21 20:32:45,702 - ControlNet - INFO - ControlNet Method inpaint_only patched.
To load target model SDXL
Begin to load 1 model
Moving model(s) has taken 2.53 seconds
27%|█████████████████████████████ | 8/30 [00:05<00:15, 1.41it/s]To load target model SDXL█████████████████████▊ | 8/30 [00:04<00:14, 1.47it/s]
Begin to load 1 model
unload clone 0
Moving model(s) has taken 2.00 seconds
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████| 30/30 [00:23<00:00, 1.30it/s]
Moving model(s) skipped. Freeing memory has taken 0.53 seconds███████████████████████████████████████████████████| 30/30 [00:22<00:00, 1.44it/s]
Total progress: 100%|████████████████████████████████████████████████████████████████████████████████████████████| 30/30 [00:24<00:00, 1.22it/s]
```

### Additional information

_No response_

Reproduce the issue from the img2img inpaint flow using ControlNet with resize mode set to “crop and resize” and a 1024×1024 resolution. Compare the masked output with the “Whole picture” inpaint-area behavior, then trace the ControlNet crop and `inpaint_only` preprocessing path. Done means the mask remains aligned and the yellow balloon is painted out as expected.

Written by the indexing model from the issue text.