Resolves #6046

Investigating the issue, I think it can be stated as two problems.

1. getpalette() returns more entries than set by putpalette().

>>> from PIL import Image
>>> im = Image.new("P", (1, 1))
>>> im.putpalette((255, 0, 0))
>>> im.getpalette()
[255, 0, 0, 1, 1, 1, 2, 2, 2, ...]

This PR solves this by adding a size member to the C ImagingPaletteInstance.
Fixing this removes the need for trimming the palette in Python in im.load().

2. Because C was not aware that a palette might have anything less than 256 entries, when converting an image to P, it considers all of the palette entries, rather than limiting itself to the palette size.

from PIL import Image
p = Image.new("P", (1, 1))
p.putpalette((255, 0, 0, 0, 0, 255))  # red and blue
im = Image.open("Tests/images/hopper.png")
im.quantize(palette=p).convert("RGB").save("out.png")

Before	After