Is your feature request related to a problem or challenge?

DataFrame.writeParquet shipped in #27, but the symmetric CSV writer is
missing. DataFusion exposes DataFrame::write_csv, which is commonly
used for ad-hoc exports.

Describe the solution you'd like

• Add a CsvWriteOptions value class covering the knobs that
  DataFrameWriteOptions + CSV-specific options expose
  (single-file output, partition columns, header, delimiter, quote,
  escape, null-value string, compression).
• Add proto/csv_write_options.proto, following the convention
  established by parquet_write_options.proto in feat(write): add DataFrame.writeParquet with ParquetWriteOptions #27.
• Expose DataFrame.writeCsv(path[, options]).
• Cover with tests in the spirit of DataFrameWriteParquetTest
  (single-file, compression, retain-after-write semantics).

Describe alternatives you've considered

COPY (SELECT * FROM df) TO '…' via SQL works but requires a
registered name and a SQL detour.

Additional context

Aligns the writer story: with this and the JSON writer, we cover the
three formats DataFusion's DataFrame API can write today
(write_parquet, write_csv, write_json).