[mkantautas]

Changes proposed in this pull request:

• On little-endian systems, resize() read and wrote I;16B and I;16N pixels in the wrong byte order when using a filter other than NEAREST. Since Support all resampling filters when resizing I;16* images #8422, _ImagingResampleHorizontal_16bpc() and _ImagingResampleVertical_16bpc() have treated I;16N as big-endian and I;16B as little-endian. They now use the same check as BoxBlur.c, Filter.c and Draw.c.
• Added a test that resizes I;16L, I;16B and I;16N images horizontally and vertically, and compares the values with the same data resized as I;16.

from PIL import Image

im = Image.new("I", (4, 1))
im.putdata([1000, 2000, 3000, 4000])
for mode in ("I;16", "I;16B"):
    resized = im.convert(mode).resize((2, 1), Image.Resampling.BILINEAR)
    print(mode, list(resized.convert("I").get_flattened_data()))

On main, this prints [1714, 3286] for I;16 and [57814, 12721] for I;16B. Big-endian 16-bit grayscale TIFF files open as I;16B, so resizing them with the default filter is affected.