feat: rewrite joins + refactoring + adding missing join-file

[SemyonSinchenko]

Close #21

It looks like it works:

Changes:

• I refactored python part to simplify the code and reduce code-duplication;
• After analyzing the joins logic I realized that the simplest way would be to pass pre-generated keys to the rust part;
• Generation of joins is slightly slower from now because of doing magic with keys in numpy on the python side;

I tried SMALL and MEDIUM generations only, I don't think that my laptop is able to handle generation of BIG.

I'm trying to realize how keys manipulation can be faster on the python side.
@MrPowers Could you please try it? I'm not very familiar with this benchmark queries, I just tried one. Anyway you are the only person who can approve it ;)

@zhuqi-lucas cc

[zhuqi-lucas]

Thank you @SemyonSinchenko !

[SemyonSinchenko]

I cannot reach @MrPowers. Theoretically, I can bypass branch protection and merge. @zhuqi-lucas how urgent is this for you?

[zhuqi-lucas]

Thank you @SemyonSinchenko for your work, i also think we can merge it, and i can try it in apache datafusion as a follow-up, and if we meet new errors, we can push another fix, do you think so?

[SemyonSinchenko]

@zhuqi-lucas Merged & Released to PyPI (0.0.3). Feel free to ping me in case of any issues!