Memory issues when generating views for multiple resource types

[bashir2]

When generating Parquet-based materialized flat views for a large dataset (MIMIC), the pipeline fails with out-of-memory issues even with a huge amount of available memory (100+ GB). This happens while each resource type can be converted individually with much less memory, so it seems there is a memory leak issues when multiple pipeline are run.

It's worth noting issues #777 and #824 where @chandrashekar-s did a lot of memory investigation and as a result we separated pipeline executions (not to be simultaneous). So this might be related or not. I have only experienced this in view generation, the original Parquet conversion (wide schema) was done successfully with all resources.