[C++] implement Statistic Schema attribute in C++

[andishgar]

Describe the enhancement requested

as follow up link Some Statistics schema which listed here has not been implemented I try to list it here.

1. ARROW:average_byte_width:exact
2. "ARROW:average_byte_width:approximate"
3. "ARROW:distinct_count:approximate"
4. "ARROW:max_byte_width:exact"
5. "ARROW:max_byte_width:approximate"
6. "ARROW:null_count:approximate"
7. "ARROW:row_count:approximate"

My focus is not just RecordBatch. These attributes have not been used anywhere

Component(s)

C++

[andishgar]

@kou Could you share your insights on which attributes should be implemented for arrow::ArrayStatistics?

[kou]

All attributes you listed should be implemented.

But can we use one issue/PR per attribute? In general, a large PR is difficult to merge.

[andishgar]

@kou, I have two questions regarding the addition of these attributes:

1- Is it necessary to modify any code on the Parquet side? If yes, could you provide a link to relevant documentation or resources for guidance? (I have no prior experience with Parquet.)

2- Do these attributes require Python bindings via Cython? If so, could this be addressed in other issues? (I lack experience with Cython and Python, but as I’ll likely need these skills for other other Apache Arrow-related issues that I 'm involved. I’d be open to starting in other issues if there’s no urgency for porting.

[kou]

1. No. I think that Parquet doesn't provide the statistics in the PR description. (@wgtmac @mapleFU Could you correct me if I'm wrong?)

2. No. We can work on the C++ implementation and the Python bindings of them as separated tasks.

[andishgar]

@kou, I think there are some inaccuracies in the Statistics Schema specification. Let me explain:

1-As you pointed out here, ARROW:row_count:approximate is not applicable to arrow::RecordBatch. However, the specification currently states that it is required:

The number of rows in the target table, record batch or array. (approximate)

2- The specification does not include any examples involving tables.

3- There's no mention of ChunkedArray. Is that because ChunkedArray is not consistently supported across all Arrow implementations?