bug: iceberg sink precompute partition key would generate wrong data if table partition key changed

[chenzl25]

In #14710, we support:

support get partition info at frontend
add the project node to compute the partition value and exchange node to shuffle by it

However, if the iceberg table partition key has been changed, iceberg sink will use the wrong partition key, or even worse it might generate data files that don't match the partition spec. Iceberg's hidden partitioning So I think we should only use the partition spec of creation to shuffle data, but when we write data to iceberg table, we should leverage iceberg's hidden partition to deal with the partition key changed problem.