fix: preserve original datatype on output instead of hardcoding signed short

[Copilot]

fslmaths -odt short (INT16) was hardcoded as the output datatype, silently corrupting UINT16 images by mapping values > 32767 to negative numbers.

Changes

• run.sh: Use fslinfo to read the original image's NIfTI datatype, then map it to the correct fslmaths -odt argument before writing the final output. Since fslmaths does not support ushort, UINT16 inputs are mapped to int (INT32), which holds all UINT16 values (0–65535) without sign corruption.

orig_dtype=$(fslinfo $nii | grep "^data_type" | awk '{print $2}')
case "$orig_dtype" in
    UINT8)   odt="char"   ;;
    INT16)   odt="short"  ;;
    UINT16)  odt="int"    ;;  # fslmaths has no ushort; INT32 holds all UINT16 values without sign corruption
    INT32)   odt="int"    ;;
    FLOAT32) odt="float"  ;;
    FLOAT64) odt="double" ;;
    *)       odt="short"  ;;
esac
fslmaths $out_unwarped $out_unwarped -odt $odt

No new dependencies — fslinfo is part of the existing FSL 5.0 install.

[akhanf]

@copilot: the problem is fslmaths does not support odt of ushort:
Error: Unknown datatype "ushort" - Possible datatypes are: char short int float double input

[Copilot]

@copilot: the problem is fslmaths does not support odt of ushort:
Error: Unknown datatype "ushort" - Possible datatypes are: char short int float double input

...

Good catch. Since fslmaths only supports char short int float double input, I've remapped UINT16 → int (INT32). INT32 can hold all UINT16 values (0–65535) without any sign corruption. Fixed in commit fix: map UINT16 to int (INT32) since fslmaths does not support ushort.

[akhanf]

cleaned up history that involved deps etc, since those dealt with in the pixi PR..
this change has been tested to work there