`ParquetRecordBatchStream` Should Return the Projected Schema

[msalib]

Describe the bug

Let's say you're trying to async read a Parquet file on S3, and that file has metadata (like "created by"). There's an inconsistency:

ParquetRecordBatchStream::schema will produce a Schema object that includes that metadata.
But ParquetRecordBatchStream will yield RecordBatches that have schema objects that don't have the metadata.

The problem is that if you create an ArrowWriter using the first schema and then try to write batches from the stream to it, the schemas won't match (the writer is expecting metadata but each batch has a schema without metadata).

Expected behavior

I'd expect that either:

• ParquetRecordBatchStream::schema produces a Schema without metadata, or
• the RecordBatches produced by ParquetRecordBatchStream have the exact same schema as what ::schema returns, or
• ArrowWriter should tolerate its supplied schema differing from the batch schemas provided to write() in metadata

[tustvold]

ArrowWriter should tolerate its supplied schema differing from the batch schemas provided to write() in metadata

#4027 relaxes the check on ArrowWriter

ParquetRecordBatchStream::schema produces a Schema without metadata, or

I think this would be consistent with ParquetRecordBatchReader, and would ensure it matches the RecordBatch that are returned. ParquetRecordBatchStreamBuilder::schema should return the schema with metadata.

[tustvold]

Updated based on #5135 (comment)

I agree this is a bug