Fix incorrect rendering when zoom-dependent text-size is used with icon-text-fit

[anandthakker]

Our current approach to icon-text-fit does not account for zoom-dependent text-size. We read the value of text-size during symbol layout:

mapbox-gl-js/src/symbol/quads.js

Lines 72 to 92 in 5bebe1c

	const iconWidth = (right - left),
	iconHeight = (bottom - top),
	size = layout['text-size'] / 24,
	textLeft = shapedText.left * size,
	textRight = shapedText.right * size,
	textTop = shapedText.top * size,
	textBottom = shapedText.bottom * size,
	textWidth = textRight - textLeft,
	textHeight = textBottom - textTop,
	padT = layout['icon-text-fit-padding'][0],
	padR = layout['icon-text-fit-padding'][1],
	padB = layout['icon-text-fit-padding'][2],
	padL = layout['icon-text-fit-padding'][3],
	offsetY = layout['icon-text-fit'] === 'width' ? (textHeight - iconHeight) * 0.5 : 0,
	offsetX = layout['icon-text-fit'] === 'height' ? (textWidth - iconWidth) * 0.5 : 0,
	width = layout['icon-text-fit'] === 'width' || layout['icon-text-fit'] === 'both' ? textWidth : iconWidth,
	height = layout['icon-text-fit'] === 'height' || layout['icon-text-fit'] === 'both' ? textHeight : iconHeight;
	tl = new Point(textLeft + offsetX - padL, textTop + offsetY - padT);
	tr = new Point(textLeft + offsetX + padR + width, textTop + offsetY - padT);
	br = new Point(textLeft + offsetX + padR + width, textTop + offsetY + padB + height);
	bl = new Point(textLeft + offsetX - padL, textTop + offsetY + padB + height);

But, at render time, we don't scale the icon to match the text size at the current zoom level. For icon-text-size: both, we could solve this by disallowing the icon-size property to be set and just using the text-size data in the icon rendering pass.

I'm not sure what we should do to handle cases where icon-text-size is used in only one dimension, and icon-size is used for the other. Some permutations of this could be handled with an extra uniform, but, for example, {icon-text-size: 'width', icon-size: [... composite expression ...], text-size: [... another composite expression ...]} would entail providing additional vertex attribute data.