Remove FromSlice in favor of From impl in upstream arrow-rs code

[alamb]

Which issue does this PR close?

Part of #1754 (see #1754 (comment))

Rationale for this change

I am trying to extract physical_plan into its own crate to improve modularity and compile times -- FromSlice is currently defined in the core crate and used in physical_plan

I am also trying to make sure the organization of the DataFusion crate is easier to understand for beginners (to make it easier to start) and thus the more we can rely on existing functionality (rather than something that is DataFusion specific) the better.

What changes are included in this PR?

Remove the DataFusion specific FromSlice trait in favor of the arrow From impl

I also added some more docs in arrow-rs to try and make them easier to use: apache/arrow-rs#4379

Are these changes tested?

Covered by existing tests

Are there any user-facing changes?

from_slice is no longer available (users should use the From impl in arrow-rs instead)

[tustvold]

As an added bonus for non-nullable primitives this conversion will now be zero-copy

[alamb]

Since this is an API change I plan to leave it open for at least 24 hours in case anyone else wants to comment

[kazuyukitanimura]

Hi @alamb,
I was using LargeBinaryArray::from_slice for performance reasons (to avoid copying). E.g.

let v = vec![vec![1u8,2,3]];
let lba = LargeBinaryArray::from_slice(&v);

Now the API is gone, I was wondering if you could advise on alternatives. I can copy the old code back into my code or I can contribute to implement From<Vec<Vec<u8>>> to arrow-rs.
I wanted to check if you may have any insight on how other users adapted this change.

Thank you in advance.

[alamb]

Hi @kazuyukitanimura

How about using https://docs.rs/arrow/latest/arrow/array/struct.GenericByteArray.html#method.from_iter_values

This works for me:

    let v = vec![vec![1u8,2,3]];
    let lba = LargeBinaryArray::from_iter_values(v);

[kazuyukitanimura]

Thank you @alamb for the reply. I somehow remembered from_iter_values was slower, but now I tested it again. Both performs similarly. Not sure where I got the wrong impression. I will go with from_iter_values. Thanks again.