[qodelicious]

Changes proposed in this pull request:

• For LA, La and PA images, getcolors() compares the whole 4 byte pixel, including the 2 unused middle bytes. Those bytes are not always the same. Image.merge() and point() leave them as 0, while new(), convert() and frombytes() copy the L value into them.
• So if an image mixes pixels from these sources, the same colour is counted more than once.

from PIL import Image

im = Image.new("LA", (2, 1), (1, 2))
bands = (Image.new("L", (1, 1), 1), Image.new("L", (1, 1), 2))
im.paste(Image.merge("LA", bands))
print(im.getcolors())  # [(1, (1, 2)), (1, (1, 2))]

• This masks out the unused bytes for two band images, the same way the unused fourth byte is already masked for three band images.