When converting, clip I;16 to be unsigned, not signed

[radarhere]

Resolves #5967

>>> from PIL import Image
>>> Image.new("I;16", (1, 1), 65535).load()[0, 0]
65535
>>> Image.new("I;16", (1, 1), 65536).load()[0, 0]
0

So we can agree that the maximum for "I;16" is 65535, a.k.a 2**16 - 1, as expected.

When converting from "I" to "I;16", there is clipping. So you would expect that clipping to limit the values of "I", the 32-bit mode, to the maximum value of "I;16", the 16-bit mode.

Except it doesn't.

>>> im = Image.new("I", (1, 1), 65535)
>>> im.convert("I;16").load()[0, 0]
32767

The following line incorrectly imagines that we need to clip to an signed range, not an unsigned range.

Pillow/src/libImaging/Convert.c

Line 40 in 38f4660

#define CLIP16(v) ((v) <= -32768 ? -32768 : (v) >= 32767 ? 32767 : (v))

This PR corrects it.