perf(parquet/pqarrow): write record ranges directly

[fallintoplace]

Summary

• Write row-group ranges directly from a record batch.
• Reuse one *arrow.Chunked wrapper per column when a batch spans row groups.
• Keep the existing path when a batch fits in one row group.
• Preserve buffered row-group boundaries, including full and zero-row cases.
• Add nested/null round-trip coverage and benchmarks.

Benchmark

Local Apple M1 Pro. -benchtime=500ms -count=3. 8 int64 columns, 8,192 rows, row-group size 256.

Method	main	change	allocs/op
Write	~2.31 ms, 4.95 MB	~2.11 ms, 4.87 MB	21,905 -> 20,761
WriteBuffered	~2.54 ms, 6.69 MB	~2.45 ms, 6.61 MB	25,407 -> 24,263

The wall-time result varies a bit between runs. The allocation reduction is consistent.

Tests

• go test ./parquet/pqarrow -run '^(TestFileWriter|TestWriteOnClosedFileWriter|TestBufferedRecWrite|TestByteArrayStatisticsStreamingReleaseBetweenBatches)$'
• go test -race ./parquet/pqarrow -run '^(TestFileWriterRangeWritesPreserveData|TestFileWriterZeroRowRecord)$'
• go vet -composites=false ./parquet/pqarrow
• go test ./... -run '^$'
• GOOS=linux GOARCH=386 go test -c -o /dev/null ./parquet/pqarrow