Add json output

[bhagenbourger]

#32

[bhagenbourger]

Hello @vianneybacoup ,
In this PR I put all test results into "target" folder excepted "output.*" because it's the default value for the output_name.
I do that to keep root folder "clean", to easier clean test results and also to avoid to add all extensions into the .gitignore file.
But, as default value is "output", I don't really know if it's really useful.
What do you think about that?

Thank you for your feedback.

[vianneybacoup]

Hello @vianneybacoup , In this PR I put all test results into "target" folder excepted "output.*" because it's the default value for the output_name. I do that to keep root folder "clean", to easier clean test results and also to avoid to add all extensions into the .gitignore file. But, as default value is "output", I don't really know if it's really useful. What do you think about that?

Thank you for your feedback.

I see that we both had the same issues, I did way too much 'rm *.parquet *.csv' past months haha
I had in mind we could have like in Python a tearDown fucntion that cleanup the output of the test automatically, but that does not exist in rust, and this is not a mandatory feature, the folder is clearly enough.
Maybe rename it with target/test_generated or something like this just to avoid a possible confusion with the other targets?

[bhagenbourger]

I moved output tests into target/test_generated/ folder.
I kept "all options" tests into target folder because they are run only by github action and generation fails if folder not exists (maybe could be a good enhancement to automatically generate intermediate folders?).
I added ctor crate to add macro to create target/test_generated/ folder before tests and clean test outputs after tests.

About json output, I did a POC using arrow (https://github.com/bhagenbourger/Fakelake/tree/poc/use_arrow_for_all_format) because I found interesting to use the same "generator" for all formats. It works but with some limitations as using the same date format for all columns. So definitively, serde-json is more flexible and I keep this implementation.

[vianneybacoup]

About arrow for json, I am also wondering if it has poorer performances than serde-json.
Fakelake whole purpose is the speed, so I'd say it should be the focus

Small adjustements to be done for this PR, thank you for the work :)

[vianneybacoup]

Thanks ! :)