Feature: Predicate pushdown

[alamb]

👋 I am just filing this ticket as I ran across some relevant research. I don't have a specific need for this yet nor did I check this repository to see if it already works

Pushing predicate evaluation into the decoder is likely to speed up certain queries substantially

https://homepages.cwi.nl/~boncz/ lists some relevant work for FastLanes

• Raufs Dunamalijevs, Universiteit van Amsterdam, Predicate Pushdown in FastLanes (2024).

[gatesn]

👋 This is sort of why I was poking at apache/datafusion#2581 to push-down all scalar expressions, not just a projection mask, and not just filters.

[alamb]

I thought apache/datafusion#2581 was about pushing down filters, but potentially rewriting them if the storage format could evaluate them more efficiently somehow

I wonder if we could add some sort of rewrite pass (maybe even in the table provider) from

cast (x as int) = 10000 

to strip te cast

I feel like getting predicate evaluation into fastlanes in general could be done without being blocked on this particular edge case. I think the filter evaluation would take most of the effort and a huge win.