Write parquet with decimal columns in load_test

[ChrisJar]

This enables using load_test to create parquet files with decimal columns as well as reading in parquet files with decimal columns for running queries. However, as the csv reader/writer doesn't support decimal yet, the decimal values have to be read in as str and cast to decimal before they're written to parquet.

Due to: rapidsai/cudf#8311 the decimal columns appear to be type object however, the underlying type is decimal and they only appear as object when calling df.dtypes due to a metadata error.

Due to: rapidsai/cudf#8354 the original precision of each decimal column is lost, but this shouldn't affect the underlying values.

CC: @randerzander