Document that histogram() uses 256 bins per channel

[radarhere]

Resolves #3958

im.histogram() groups pixel counts into 256 bins per channel, even for I and F mode images (which have pixel values of 0 to 65535, meaning that you could have 65536 bins in theory).

from PIL import Image
for mode in ("1", "L", "I", "F"):
    im = Image.new(mode, (1, 1))
    assert len(im.histogram()) == 256

The ImageStat module performs operations on that histogram. This means that for I and F mode images, ImageStat can only report what it has been given.

from PIL import Image, ImageStat
im = Image.new("I", (100, 1))
px = im.load()
for i in range(99):
    px[i, 0] = 65535 - i
stat = ImageStat.Stat(im)
print(stat.mean[0])  # 251.47

So I've added a note that mean, median,rms and the maximum extrema can't be more than 255.