Never just toss out ideas, keep them on paper and doodle, or just take half an hour in hammer and slam something out. Often you'll come up with the next part after you see the idea in a different medium.
If you play a single gamemode alot you can already predict ways on how people are going to handle it. By this info you can already scrap things that dont work. There is no need to build them since if the idea partialy works, you prob will already think about ways to counter it. And if that doesnt work yet you still are convinced it works than you build it and verify your idea, and fix it afterward. Sometimes it requires a playtest, but many times its not even needed.
Also, dont forget that a quick 2d drawing on paper describing the idea can indicate enough. You dont need a perfect scale on paper, but if you draw it you can spot fast if it fits or not, and if it doesnt fit, where thats going to be.
Scrapping early can save alot of time. And if you scrap bad ideas you wont clutter them between the good ideas.
Ofcourse, if you just got a small section you cant use yet then drawing/building it for later use isnt a bad idea. But you should still be able to test it without building it to begin with.
I would recommend keeping a theme throughout development.
That realy depends on what you want to make. For skullcove the original idea was maya themed. It makes the dev shape at least have some distinctive areas. But other than that it never was used. For intercept it was space theme (even though the only reference point was the dish at that time), that became egypt.
Being able to swap styles can do alot, not every shape works for every map. Going hong kong style using small geometry is for example something that isnt going to look convincing. Going desert instead could be a far more efficient choise. Dont restrict your map too heavily on a style. Gameplay should be first.