Further Reduced File System Operations in Experimentation API

If we work on a shared drive, the file system operations are the most likely operations to cause problems.
With this release, we further reduce them a bit.
We now remember which directories already exist and also directly try to create files from which we do not know whether they exist without first checking for their existence.
This should reduce the number of file system operations per log file by 2.
As a result, the experiment execution should become a bit more reliable.
It should already be quite reliable if you work on your local PC or if you have few involved processes or a fast connection to the shared drive.
But if I run big experiments with many threads, I sometimes get I/O errors and thus, automatic restarts.
With this release, we try to further reduce those.