since set was already problematic, leaving this might solve the latest-macos 3.7 issue. the problem is with python itself, not with the pipe protocol what i've proven, so it's safe to do sth like this imo
I've simplified the code for tagger's pipe server, i've added better logging for pipe, I've replaced dicts with strings to fix python's problems with the tests. Pipe works fine, the problem seems deeper than we thought, this approach also tests the pipe correctly.
All the issues with Unix tests was caused by the BrokenPipeError which is actually hard to reproduce. I've added a basic handle procedure: re-create the pipe and log a warning.
Should work now, and if not, we at least have more verbose logging system
I've decided to remove the filename tests becaue they provide no useful value during the tests. Everything has to be mocked and so on... Debug-logging the pipe path should do, because during the tests we cannot be 100% sure that the specific folder will be writable, the env variable will be declared etc.
