Enhancement: Improve NDWI stability and clean redundant imports in create_masks.py

[LJitendra05]

Summary

This issue proposes several small improvements to data_preprocessing/create_masks.py to improve code quality and numerical stability during NDWI computation.

Proposed Improvements

1. Remove duplicate imports

Currently the file contains redundant imports such as:

from rasterio import mask
from rasterio.mask import mask

and duplicated os imports.

Cleaning these improves readability and avoids confusion.

---

2. Improve NDWI numerical stability

The NDWI calculation currently uses:

ndwi = (green - nir) / (green + nir)

If (green + nir) == 0, this can cause division-by-zero warnings and generate invalid values.

Suggested improvement:

ndwi = (green - nir) / (green + nir + 1e-8)

This avoids numerical instability while keeping NDWI values in the same range.

---

3. Improve NDWI scaling to 8-bit

Currently NDWI values are converted using:

(out_image * 127) + 128

Since NDWI ranges from [-1, 1], a clearer scaling approach would be:

((ndwi + 1) * 127.5).astype(np.uint8)

This explicitly maps the NDWI range [-1,1] → [0,255].

---

Benefits

• Improves numerical robustness of NDWI computation
• Reduces redundant imports
• Clarifies NDWI scaling logic
• Improves maintainability of preprocessing code

---

If this enhancement is acceptable, I would be happy to submit a pull request implementing these improvements.