GH-45034: [C++] Remove Parquet requirement from Arrow Acero and from Arrow Dataset when not necessary

[raulcd]

Rationale for this change

Currently we are defining Parquet as a requirement for Acero and for Dataset on ArrowAceroConfig.cmake.in and ArrowDatasetConfig.cmake.in respectively.

Parquet should never be necessary for Acero and could be necessary for Dataset only if it has been built with Parquet support.

What changes are included in this PR?

Remove Parquet from ArrowAceroConfig.cmake.in and add it as optional dependency for ArrowDatasetConfig.cmake.in

Are these changes tested?

I've tested with a really minimal build of Arrow and Pyarrow. I don't think we want to add a build without Parquet but with Dataset and/or Acero at CI just to test this scenario.
Logs for the local build without Parquet:

-- Found Cython version: 3.0.11
-- Arrow version: 19.0.0
-- Found the Arrow shared library: /home/raulcd/code/dist/lib/libarrow.so.1900.0.0
-- Found the Arrow import library: ARROW_IMPORT_LIB-NOTFOUND
-- Found the Arrow static library: 
-- Building PyArrow with Dataset
-- ArrowAcero version: 19.0.0
-- Found the ArrowAcero shared library: /home/raulcd/code/dist/lib/libarrow_acero.so.1900.0.0
-- Found the ArrowAcero import library: ARROW_ACERO_IMPORT_LIB-NOTFOUND
-- Found the ArrowAcero static library: 
-- ArrowDataset version: 19.0.0
-- Found the ArrowDataset shared library: /home/raulcd/code/dist/lib/libarrow_dataset.so.1900.0.0
-- Found the ArrowDataset import library: ARROW_DATASET_IMPORT_LIB-NOTFOUND
-- Found the ArrowDataset static library: 
-- Building PyArrow with Acero
-- Parquet Encryption is NOT Enabled

The IMPORT_LIB-NOTFOUND has always been present and it's not new.

Are there any user-facing changes?

No as this should have already been the case.

• GitHub Issue: [Python][C++] pyarrow does not build without parquet #45034

[github-actions]

⚠️ GitHub issue #45034 has been automatically assigned in GitHub to PR creator.

[kou]

+1