Add pl.concat_arr: similar to pl.concat_list but returns an array data type

[mkleinbort-ic]

Description

Since the creation of the pl.Array data type it is a good way to approach "list" columns where all the lists have equal lenghts.

The most common way I see this happening is when one does a pl.concat_list('col_1', 'col_2', ..., 'col_n') to get an column where all lists have n elements.

One can do pl.concat_list('col_1', 'col_2', ..., 'col_n').list.to_array(width=n) but that requires knowledge of n (which might be dynamic when using selectors instead of explicit column names)

Moreover, the list format is (probably) not as optimized as the array format for this usecase.

So... how about a pl.concat_arr (or pl.concat_array) to leverage these benefits?

[mkleinbort-ic]

It came to mind because some code I'm working on had the issue that

    .select(y_pred_samples = pl.struct(pl.all())) # Combine into a struct column

was 1min faster than

    .select(y_pred_samples = pl.concat_list(pl.all()).list.to_array(width=100)) # Combine into an array column

but at the end of the day I'm just going to be working with min/max/mean of the array values

[deanm0000]

I think having a means to create an array is good but I don't think the timing is relevant. Creating a struct is as close to free as you can get since it doesn't have to check anything but making a list or array requires it to ensure the data types are compatible for each row.

[mkleinbort-ic]

Tbh I'm still waiting to see the benefits of the pl.Array data type. I love structs, but it's not easy to do operations on the values (.min(), .max(), .mean(), etc...)