Fixed bug in decoding large images

[radarhere]

Resolves #1475

[hugovk]

#1475 (comment) suggests changing all ints to py_ssize_t.

What are the pros and cons of using unsigned int over py_ssize_t?

[radarhere]

It didn't occur to me that py_ssize_t was also applicable here, thanks for spotting. I've changed the commit to use that.

I've also changed the other decoders as well, to fix a compilation warning.

[elhuhdron]

Many PIL operations are working for me for image manipulations of images over 32 bit in size, which is fantastic. However, I still have a problem with fromarray even after this fix (tested using PIL version 6.1.0 and numpy version 1.16.4):

The test case is relatively simple:

from PIL import Image
import numpy as np

# this works
a = np.zeros((46340, 46340), dtype=np.uint8)
b = Image.fromarray(a)
del a,b

# this works
b = Image.new('L', (46341, 46341))
del b

# this does not work, fails with:
# MemoryError: Integer overflow in ysize
a = np.zeros((46341, 46341), dtype=np.uint8)
b = Image.fromarray(a)

This is likely because 46341*46341 > 2^31-1
Is this easily fixable?

[radarhere]

@elhuhdron could you open a new issue with your problem?