[FEATURE] Projection Pushdown (Parquet)

[joocer]

Push down the projection to the read step.

This should improve performance by handling less data in the processing steps.

[joocer]

only implement for the external readers (blob, nosql and sql readers)

[joocer]

This is probably going to be harder than just collecting all the tokens which are labelled as identifiers, especially when there's joins or sub queries, or aliases.

[joocer]

Implement hint NO_PUSH_PROJECTION at the same time

[joocer]

Do this as a first page to gather all the fields, then intersect with the selected fields and use that on future page reads.

This will mean no benefit on small datasets (single page), but that's not what would benefit from this anyway.

A '*' in the field list should disable the optimization.

This should be reflected in EXPLAIN.

Note that NATURAL JOIN should add a '*' to the field list when implemented.

This should NOT be the same approach taken for other data types.

[joocer]

This may conflict with the schema evolution feature, what happens if we select a column that doesn't exist. Maybe we need to wrap in a try and do more expensive work if it fails.

[joocer]

Can the field list be converted to a set earlier and once

Can we use the schema to update the field list set