Refactor partial parsing cache to be per dbt project

[tatiana]

In #904, we started caching the partial parsing and manifest files. However, that happened at a DbtDag and DbtTaskGroup level. By then, we considered configuring at the dbt project level, but there were concerns regarding concurrent tasks running in the same Airflow worker node.

This ticket aims to refactor the caching to be set at a dbt project level. We should consider using the _safe_copy method created in #1086. We should create alphas and have as many users as possible trying out this.

This refactor will improve with filling up users' disk: #1042