Fix Spark cache key to include hadoop-profile #1799
Description
In the CI workflow tpcds-reusable.yml, current Spark cache key only includes sparkver and scalaver.
This causes a defect: different hadoop-profile values download different Spark tarballs but share the same cache, leading to pollution or wrong binaries.