Reduce overall io wait time for parquet files

[scsmithr]

Describe the bug

Running the query select * from 's3://glaredb-public/userdata0.parquet'.

Fetching metadata/content length: ~400ms
Parquet streaming: ~1000ms

To Reproduce

select * from 's3://glaredb-public/userdata0.parquet'

Expected behavior

Faster