Prune casted partition columns

[sopel39]

Some tables have partition keys based on strings. Then users might write queries like:

WHERE CAST(part_col AS DATE) > DATE '1992-01-01'

TupleDomains are derived from such predicates. However, connectors don’t really consume CAST(part_col AS DATE) > DATE '1992-01-01' predicate and such predicate remains in ScanFilterProject as remaining predicate.

This significantly reduces cache hit ratio as part_col predicates are often unique.

Hence, in order to improve subquery cache for such cases we need to:

1. Introduce new method like: TupleDomain<ColumnHandle> ConnectorPageSourceProvider#getSplitPredicate, which would return TupleDomain<ColumnHandle> that would describe split. Such method can actually be used internally by prunePredicate so it feels like it fits in place. This method in particular will return partition_col value as a Domain#singleValue

2. Enhance CommonPlanAdaptation.PlanSignatureWithPredicate to also contain predicates that couldn’t be translated into TupleDomain but touch scan columns, e.g: CAST(part_col AS DATE) > DATE '1992-01-01'

3. Enhance CacheDriverFactory to use ExpressionInterpreter on non-TupleDomain predicates with NullableValues from getSplitPredicate. This way such predicates could be simplified.

4. Put simplified non-TupleDomain predicates into CacheSplitId