Subsume non-identity partition predicate in Iceberg

[findepi]

Before the change, Iceberg connector accepted all predicates expressible
as TupleDomain on primitive columns and they were used to filter data
files during split generation. However, only predicates defined on
identity partitioning columns were subsumed into connector.

This commit extends Iceberg capabilities to subsume predicates on
partitioning columns. Besides subsuming predicates on identity
partitioning columns, it also subsumes predicates if they align with
partitioning boundaries. For example, for truncate(col, 2) (round to
100s) partitioning, predicates col >= 1200 OR col > 1199 are
subsumed, while col > 1200 or col > 1250 are not.

This change is especially important for Iceberg OPTIMIZE table
procedure, which requires the WHERE condition to be fully subsumed
into the connector. It is also helpful for DELETE, as it allows to do
metadata-only delete in more cases, where we don't really needing to do
a row-level delete.

Fixes #7905
For #12362

[findepi]

Still a draft, so just FYI @alexjo2144 @losipiuk @ebyhr @erichwang @phd3 @findinpath @homar

[findepi]

CI #12726

[findepi]

This is ready to review.
Test-only prefix is extracted to: #12865

@alexjo2144 @losipiuk @ebyhr @hashhar @erichwang @martint PTAL

[alexjo2144]

Neat way of implementing this. I want to give it another read

[erichwang]

Mostly nits here, but the logic looks sound.

[erichwang]

if you follow the java Math library naming conventions, this would be called: previousBefore or nextDown.

[findepi]

I didn't mean to follow the java Math library here.
Which particular methods are you referring to?
("nextDown" still sounds odd to me)

anyway, i think this should be provided by the Type itself, so let's discuss best naming in #12797

[losipiuk]

looks good overall

[findepi]

AC

[findepi]

Build is green. Rebasing after #12865 merged.