GH-43135: [R] Change the binary type mapping to blob::blob

[eitsupi]

Rationale for this change

Many packages, including nanoarrow, use the blob class provided by the blob package to represent a vector of binary data.
However, this package did not use blob.
Therefore, when this package convert Arrow Binary type to R, it was difficult for other packages to interpret it as a Binary type.

What changes are included in this PR?

The current special processing for the arrow_binary class should also be performed for the blob class.
Also, arrow_binary, arrow_large_binary, and arrow_fixed_size_binary will be changed to subclasses of the blob class.

Are these changes tested?

Rewrote some existing tests to use blob instead of arrow_binary.

Are there any user-facing changes?

Yes.

But the traditional classes have been changed to subclasses of blob, so little impact is expected.

• GitHub Issue: [R] Change the binary type mapping to blob::blob #43135

[github-actions]

⚠️ GitHub issue #43135 has been automatically assigned in GitHub to PR creator.

[pitrou]

@paleolimbot @assignUser Does one of you want to take a look?

[assignUser]

Sounds like a good idea to improve compatibility with the larger ecosystem. Especially if nanoarrow already does it.

Not sure if there were reasons to not implement this historically @jonkeane @thisisnic ?

[paleolimbot]

Sounds like a good idea to improve compatibility with the larger ecosystem. Especially if nanoarrow already does it.

+1! I am not aware of a practical reason to differentiate between fixed size/large/regular binary at the R level (which seems to be the reason noted in the issue thread) but perhaps others will add that context.

[jonkeane]

Thanks for this

[eitsupi]

Thank you. I opened #45709 as a follow-up issue.

If a version is released that includes this, we can work on #45709 in the next release after that.

[eitsupi]

Can anyone merge this? Thanks!

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit 713d09b.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 4 possible false positives for unstable benchmarks that are known to sometimes produce them.