Resolves #3977

That issue attempts to draw two halves of a word separately, expecting them to occupy the same pixels as the complete word. The example given there is "par" and "adise" vs "paradise".

#4959 helped by adding textlength, but did not resolve the issue completely. After switching to textlength, there was still a difference.

The ImageDraw text operation essentially has two parts - it generates a mask from the text characters with no awareness of the text position, and then pastes that mask onto the image at the relevant position. I had a thought - what if the position did matter, and text characters were rendered differently if they started halfway through a pixel?

With this PR, I have added a start argument to getmask and getmask2.

from PIL import ImageDraw, Image, ImageFont

img = Image.new('RGBA', (120, 80), color=(0, 0, 0))
draw = ImageDraw.Draw(img)
font = ImageFont.truetype('/System/Library/Fonts/Supplemental/Arial.ttf', size=48)

mask = font.getmask2("adise", "L", start=(0, 0))[0]
draw.draw.draw_bitmap((0, 2), mask, 0)

mask = font.getmask2("adise", "L", start=(0.4, 0))[0]
draw.draw.draw_bitmap((0, 42), mask, 0)

img.save('example.png')

produces "a", "d", "s" and "e" in the same location, but a different "i" for the first line compared to the second.

Changing ImageDraw's text method to pass through the fractional coordinates of the text resolves the issue, allowing "par" and "adise" to fill the same pixels as "paradise".

To try and future proof the API, I have set start to be a tuple, so that both horizontal and vertical starting positions can be specified.