Perf improvements

Some performance improvements, mostly as discussed in #99 (closed).

This reduced the CLI-runtime for a synthetic dataset with 1000000 rows and 20 columns range tested on every column from ~103 to ~30 seconds as measured with the linux time-utility.

As the masking is heavily under tested, please thoroughly review these changes @palmb and @luenensc !

Not sure, why the Pipeline fails as it runs on my machine. I have to dig into that...