ARROW-17932: [C++] Implement streaming RecordBatchReader for JSON

[benibus]

See: ARROW-17932.

Adds a json::StreamingReader class (modeled after csv::StreamingReader) with an async-reentrant interface and support for parallel block decoding.

Some parts of the existing TableReader implementation have been refactored to utilize the new facilities.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-17932

[benibus]

A couple notes on this:

• I decided to hold off on documenting threading/reentrancy characteristics until I could get a second opinion, as this was my first exposure to the project's async facilities. As it's written, the stated limitations of the CSV equivalent should still apply (no async-reentrancy, no multi-threading). However, here, it seems possible to launch the parsing/decoding tasks in parallel once a chunked block has been serially generated. There may be complexities I'm not considering though.
• Test coverage handles the common cases, but likely isn't comprehensive enough - although some of it should be handled by the existing TableReader tests since it uses common code.

[pitrou]

Some answers to the points above:

• It seems like this may be async-reentrant (and theoretically could be, as you point out)

• Tests should indeed be complemented with: 1) tests for errors 2) tests with multiple input blocks 3) stress tests, especially async-reentrant calls to ReadNextAsync (MakeReadaheadGenerator may help with that)

[wjones127]

I'm still learning the async utilities in Arrow, and this PR was a very helpful read. 😄 Had two questions on it.

[pitrou]

That was an interesting read! The general structure looks sound, here are some comments.

[pitrou]

By the way, can you also:

1. rebase/merge from master
2. add documentation for the new class in https://github.com/apache/arrow/blob/master/docs/source/cpp/json.rst and https://github.com/apache/arrow/blob/master/docs/source/cpp/api/formats.rst#line-separated-json
   ?

[benibus]

@pitrou I still have to do the docs, but feel free to bring up any further points on the code in the meantime.

[benibus]

I just opened a small PR that addresses the chunker issue: #14843. If that gets merged, I have a test I can add for the newlines_in_values = true case (which would crash otherwise). Then, this will be ready for re-review.

[wjones127]

Thanks for adding docs!

[pitrou]

Thanks for the update!

[benibus]

@pitrou Just pushed a fix.

Also, not sure if my rationale on the std::remove table validation stuff from yesterday is sound - but if not, I can change that too.

[pitrou]

This PR looks really good now. Great work @benibus !

[ursabot]

Benchmark runs are scheduled for baseline = 375372b and contender = fe5c9fe. fe5c9fe is a master commit associated with this PR. Results will be available as each benchmark for each run completes.
Conbench compare runs links:
[Finished ⬇️0.0% ⬆️0.0%] ec2-t3-xlarge-us-east-2
[Finished ⬇️0.13% ⬆️0.23%] test-mac-arm
[Finished ⬇️1.36% ⬆️4.62%] ursa-i9-9960x
[Finished ⬇️14.03% ⬆️9.62%] ursa-thinkcentre-m75q
Buildkite builds:
[Finished] fe5c9fe3 ec2-t3-xlarge-us-east-2
[Finished] fe5c9fe3 test-mac-arm
[Finished] fe5c9fe3 ursa-i9-9960x
[Finished] fe5c9fe3 ursa-thinkcentre-m75q
[Finished] 375372b7 ec2-t3-xlarge-us-east-2
[Finished] 375372b7 test-mac-arm
[Finished] 375372b7 ursa-i9-9960x
[Finished] 375372b7 ursa-thinkcentre-m75q
Supported benchmarks:
ec2-t3-xlarge-us-east-2: Supported benchmark langs: Python, R. Runs only benchmarks with cloud = True
test-mac-arm: Supported benchmark langs: C++, Python, R
ursa-i9-9960x: Supported benchmark langs: Python, R, JavaScript
ursa-thinkcentre-m75q: Supported benchmark langs: C++, Java

[ursabot]

['Python', 'R'] benchmarks have high level of regressions.
ursa-i9-9960x