Do not render text if image has zero width or height

[radarhere]

In FreeTypeFont.getmask2(), only spend time rendering text if the image it will be rendered onto has a non-zero width or height.

[hugovk]

Seems fine, but I wonder how often someone will draw on an image with non-zero width/height, and how much time this would actually save them?

[radarhere]

getmask2() isn't actually rendering onto a user-supplied image. It is creating an image from the dimensions returned by font.getsize

Pillow/src/PIL/ImageFont.py

Lines 761 to 769 in 80edcd1

	size, offset = self.font.getsize(
	text, mode, direction, features, language, anchor
	)
	if start is None:
	start = (0, 0)
	size = tuple(math.ceil(size[i] + stroke_width * 2 + start[i]) for i in range(2))
	offset = offset[0] - stroke_width, offset[1] - stroke_width
	Image._decompression_bomb_check(size)
	im = fill("RGBA" if mode == "RGBA" else "L", size, 0)

[nulano]

Sure, but this would only happen if a user tries to render a string containing spaces only, right?

In which case I'd argue that might be up to the user to prevent, as that doesn't seem like something that would be commonly done.

[radarhere]

Aside from performance motivations, I was primarily thinking about https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=48708

Locally, #6842 resolved the problem with memory usage, but apparently oss-fuzz itself doesn't think it is fixed yet, so this is a second attempt. The font in the issue can trigger an OSError: raster overflow. This change avoids that.