[core][flink] supports count star pushdown for all-global-indexed filters.

[steFaiz]

Purpose

This introduce a PredicateVisitor to decide whether a Predicate is all covered by scalar index (e.g. btree now).
Then we can accelerate count(*) queries on indexed columns without touching data files. This is useful in some AI situations. For example, before filtering high quality data, we might want to check the portion to decide the threshold.

Tests

org.apache.paimon.globalindex.ScalarIndexedFieldsVisitorTest for UnitTest
org.apache.paimon.flink.BatchFileStoreITCase for ITCase

[steFaiz]

I'm not sure whether it's safe. The result may be inaccurate. This PR assumes that: If the whole data predicates can be covered by global index, those predicates can be fully consumed by paimon. But actually this is not right, because the global index might be not covering all data. Moreover, current Index is not exact, for example for unsupported queries e.g. LIKE, NOT_LIKE, the index reader will return ALL_NON_NULL_ROWS.

[steFaiz]

I think an alternative approach is to introduce a new procedure or a special option.