core: snappy as default for parquet compession

[rbalamohan]

PR to use snappy as default compression for parquet instead of gzip. #5658

[sumeetgajjar]

org.apache.iceberg.TestSplitScan > test[format = parquet] FAILED

Hi @rbalamohan, the above test is failing due to the change in compression codec, the size of the data file is now ~ 77MB and with a 16MB split size, it results in 5 scan tasks.

iceberg/data/src/test/java/org/apache/iceberg/TestSplitScan.java

Line 97 in 97c85aa

int numRecords = 2500000;

Reducing the number of records to 2000000 to get the same file size when gzip was used resolves the test failure.

[rdblue]

From the fairly broad testing that I've done, snappy is never a good choice for compression. This choice depends on what you want to optimize. Snappy is often fast, but gets very poor compression rates. LZ4 is a much better choice if you're optimizing for write speed because it is usually faster and smaller than snappy. But if you're optimizing for compression, I probably wouldn't choose LZ4.