Parquet scans still set up a reader for files whose row groups are all pruned #26475

Describe the bug

A selective filter over Parquet files still pays to set up a reader for every file range, even when row-group statistics or Bloom filters have already ruled out every row group in it. On wide tables that setup grows with the column count, so a query that reads nothing from most files still spends CPU opening each one.

To Reproduce

I've got a PR for this so I'm not going heavy into that, and well it's a relatively obvious change.

Expected behavior

No response

Additional context

No response

Willingness to contribute

None