pass drug_descriptors (contains mordred descriptors) to consecutive build runs for datasets

[ymahlich]

This feature / fix serves two purposes:

• fixes potential floating point operation inconsistencies that lead to the behavior observed in discrepancies in mordred values for the same improve_drug_id across different datasets #321 (discrepancies in modred values for the same chemical compound (calculated form identical SMILES strings) in different drug response datasets
• speeds up the build process

[sgosline]

This is a much simpler fix than trying to figure out the cause. Update build_drugs.sh to pass drug descriptor files in addition to drug file.

[sgosline]

We tried rounding, it did not work.