Improve data loading for pgbench tests

We currently do the following:

- Create a local table & indexes
- Load data
- create distributed table

The third step is slow because we're actually loading the data on the shards again, with indexes already created. This is especially becoming slow if the data size relatively high.

Instead, we can use the functions [here](https://gist.github.com/onderkalaci/89d83fc14b1c8bfc378a9bab9d7d2be5), which skips (a) loading the data on the local table (b) creating the indexes after the data load on the distributed tables