"import" and const calculations are compile-time activities, whereas init() and global var assignment are run-time activities.
How does the time to compile your code to an executable, compare to running the tests? If compiling the code is fast but running the tests is slow, that rules out any problems with "import".
If it takes "dozens of seconds" to compile your code, that doesn't sound right unless either your codebase is huge, or you're running in a very slow environment (e.g. virtualization without hardware support like VT-x).
If it takes "dozens of seconds" to start your test suite, that doesn't sound right unless you're accidentally compiling and testing all your dependent libraries as well as your own code. Or it could be that your test *is* starting quickly, but the first test runs very slowly - maybe it's opening network connections or doing DNS resolutions which time out, or something like that.
One thing you can try is to use strace to see system calls:
strace -f -s128 go test ...etc...
This might give you some clues, especially if you see a long pause in the strace output (what system call did it do just before the pause?)
Another idea is to do a binary chop. First, make a copy of your project and remove all the tests except for one "hello world" test; see if it's fast. If it is, that shows that it's either building or running your own test code which is slow, not the importing of third-party libraries. Then do the same but cut out only the first half of the tests, then repeat cutting out only the second half of the tests. When you've found which half is slow, cut that in half again - and so on.