If you want to prove the maximum stack usage of a recursive evaluation, you need to prove how deep it goes. Since that would depend on the inputs, you need to statically determine the limits on the inputs. It's all possible, but cases where it applies would be rare.

Since with recursion you can have a large but finite number of calls that could blow the stack, but not have the same problem with a for loop.

Yes, you have to account for that. Also, tail recursion should be fine, if your compiler is smart enough. (I guess since the program is looking at assembly, it couldn't tell optimized tail recursion from a loop.)