You're assuming they didn't arrive at these settings for borders, weight, classes of label via A/B testing of various alternatives. Google has enough resources available that for something like this it could easily substitute brute-force testing for "best-practice" design (and there's a number of arguments why that might be the better choice, including the fact that design practice seems to evolve over time.)
>They would still have to think of these ideas in order to test them.
Text has background, envelope, whitespace (leading, kerning, etc.), font, outline, fill (and probably other general properties too). It's not A/B testing but I could see engineers iterating over variations - should the envelope follow the curves of the text or be a box or blob, etc.. All spacing, be it kerning or border-width can be iterated over, colours too - indeed thinking of this makes me want to write an Inkscape plugin to help in font choice.
You can level your objection again but I don't think one would call thinking what shape an outline could be "design" per se. Indeed if you're just presenting alternatives to a user and looking for the one with the best feedback I think you arguably have done away with a design stage.
Well, if you're in a lab, you do a test where you say "I'm going to show you a map. Click on New York as soon as you can." etc. If you want to be more sophisticated, you add in eye tracking and see how much "scanning" people do before finding the target. I'm sure Google has both techniques available in-house.
And I'm sure if you were a little more clever you could find ways to do a similar test on normal google maps users without them knowing it.