vectorize layout elements

[badGarnet]

This PR adds two new vectorized page level dataclasses:

• TextRegions: replaces a list[TextRegion] and store coordinates and texts as numpy arrays for more efficient memory operations when the number of items is large
• LayoutElements: replaces a list[LayoutElement] and store data in numpy arrays as above
  In addition this PR refactors yolox model inference to use those two new classes above internally while keeping the list data structure still available for backward compatibility (e.g., passing into a PageLayout object).

test

compare memory usage and runtime on a pdf image using

from unstructured_inference.inference.layout import process_file_with_model


def main():
    f = "/Users/yaoyou/Downloads/002489.pdf"
    layout = process_file_with_model(f, model_name="yolox")
    # replace elements_array with elements when using main branch
    print(f"fount {len(layout.pages[0].elements_array)} elements")


if __name__ == "__main__":
    main()

The peak memory is smaller on this branch (exact amount depends on the number of layout elements detected) and processing time is slightly faster (since this PR skips generation of list of LayoutElement from numpy array output of the yolox model).