Shorter row count threshold of parallel

[pan3793]

There is a hard-coded 1 million row counts threshold of Parallel, which is not friendly for distributed computing engines to generate data in parallel.

https://github.com/trinodb/tpcds/blob/8a02abbba864feedc2afd078c8153d66a95bb2d4/src/main/java/io/trino/tpcds/Parallel.java#L26-L36

For example, in Spark, generate tpcds.sf1.web_sales in single thread cost 23s.

WDYT to set the threshold to 1k or 10k instead of 1m?

[pan3793]

cc @ebyhr