Use ex unit for measure values

[thisisdano]

I'm finding that the ch unit is a poor unit for doing our measure calculation, since the 0 character on which it's based has significant variance from font to font. Since we're already normalizing our font size in such a way that fonts display with a similar x-height at each grade in our font-size system, we should use the ex unit for a more consistent and reliable output. The ex unit has universal support.

This is also distorting our measure output tokens in the direction of being generally too wide — far outside the 50-75 target character count in our guidance. That said, the guidance is maybe a little bit skewed narrow, as a popular readable site like medium.com has a line length of around 75-78 characters, and we could a bit longer than that for something like a small caption.

I'm also proposing that we add an additional token to accommodate some of gamut we'll lose by concentrating our coverage on the guidance range.

token	desired output	proposed value	current value	current output in Public Sans
1	~50 characters	44ex	40ch	51 characters
2	~66 characters	60ex	60ch	80 characters
3	~72 characters	64ex	66ch	88 characters
4	~78 characters	68ex	72ch	95 characters
5	~86 characters	74ex	77ch	100 characters
6	~96 characters	88ex	—	—

Beta 6	Beta 7
'1'	'1'
'2'	'4'
'3'	'5'
'4'	'6'
'5'	'6'

[maya]

Done in #2964.