arrow

ARROW-9782: [C++][Dataset] More configurable Dataset writing

Python:
- ParquetFileFormat.write_options has been removed
- Added classes {,Parquet,Ipc}FileWriteOptions
- FileWriteOptions are constructed using FileFormat.make_write_options(...)
- FileWriteOptions are passed as a parameter to _filesystemdataset_write()

R:
- FileWriteOptions$create(...) to make write options; no subclasses exposed in R
- A filter() on the dataset is applied to restrict written rows.

C++:
- FileSystemDataset::Write's parameters have been consolidated into
  - A Scanner, from which the batches to be written are pulled
  - A FileSystemDatasetWriteOptions, which is an options struct specifying
    - destination filesystem
    - base directory
    - partitioning
    - basenames (via a string template, ex "dat_{i}.feather")
    - format specific write options
- Format specific write options are represented using the FileWriteOptions hierarchy. An instance of these can be constructed from a format using FileFormat::DefaultWriteOptions(), after which the instance can be modified.
  - ParquetFileFormat::{writer_properties, arrow_writer_properties} have been moved to ParquetFileWriteOptions, an implementation of FileWriteOptions.

Internal C++:
- Individual files can now be incrementally written using a FileWriter, constructible from a format using FileFormat::MakeWriter
- FileSystemDataset::Write now parallelizes across scan tasks rather than fragments, so there will be no difference in performance for different arrangements of tables/batches/lists of tables and batches when writing from memory
- FileSystemDataset::Write::WriteQueue provides a threadsafe channel for batches awaiting write, allowing threads to produce batches as another thread flushes the queue to disk.

Closes #8305 from bkietz/9782-more-configurable-writing

Lead-authored-by: Benjamin Kietzman <bengilgit@gmail.com>
Co-authored-by: Neal Richardson <neal.p.richardson@gmail.com>
Signed-off-by: Benjamin Kietzman <bengilgit@gmail.com>