Rescaling segmentation issue causing failures in ~15% of cases

### What happened?

In about 15% percent of cases, especially when the participant is 10 years or older, there appears to a registration issue in which the segmentation does not seem to be scaling back to the input image (see image)

First guess is that this is due to bad flirt registrations, but where?
This image is the input image, the working directory ../cropped/T1w/sub-105097100_ses-4_0000.nii.gz, and the infant_MNI_t1 template, none of which seem to be right.

If this IS just a registration error, and IF more frequent in older children and adolescents segmented by bibsnet, then it might make sense to either a) provide a target with more similar image characteristics appropriate for older participants in the space and scale of the infant MRI, or b) use a more robust registration method (for example, my current favorite, synthmorph in FreeSurfer.

If it is a matter of failing to reapply the inverse transform in some cases, that might be a more subtle bug to sus out. Please let me know if I can give access to a reproducible sample, without being available to the world.

### What command did you use?

```shell
I am using docker image dcanumn/bibsnet:release-3.6.0
docker run --gpus all -u $(id -u):$(id -g) --rm -v /mnt/friar/segproj/test:/input -v /mnt/friar/segproj/testout/:/output dcanumn/bibsnet:release-3.6.0 /input /output participant -v
```

### What version of BIBSnet are you using?

3.6.0

### Directory Structure

_No response_

### Relevant log output

```shell

```

### Add any additional information or context about the problem here.

_No response_