fix: empty CIBW_BUILD should still count as "*"

[henryiii]

Mentioned in #587 - currently, there's an asymmetry between not setting CIBW_BUILD, and setting it to "", that can be surprising, especially in a matrix. This treats an empty string the same way as being undefined.

[henryiii]

Ouch, tests are pulling from latest? Manylinux2010 tests will need to stop trying to do that for Py2 for latest.

[henryiii]

Tests are dead until #591 goes in.

[YannickJadoul]

Tests are dead until #591 goes in.

Sorry, yeah. I'd just noticed :-)

[henryiii]

PS: I also am somewhere between disliking and vehemently hating the a, b = c, d syntax. You'll notice the long block of code in mplhep I was refactoring had tons of it. I don't find it any better than writing a = c; b = d;, which is a fake way to compress lines, losing readability. And if you are trying to scan the code base for b =, it hides this by being the only construct in Python (except for the walrus operator, in 3.8+) that does have the assignment as the first item in a line. For a, b = c or a, b = b, a, you are doing something explicit (tuple unpacking or variable swapping), so losing first item assignment is a valid price to pay. But not for assignment grouping - a blank line does that much better.

(I'm planning to add do/do not's to Level Up Your Python, and this is one I'm planning to add. Mentioning it here gives the option for rebuttals before I add it. :) )