[core] Use a single BTree scan for bounded range predicates

[jerry-024]

Purpose

A filter such as duration >= 30 AND duration < 120 currently evaluates the two BTree predicates separately, scans both half-ranges, and intersects their candidate row IDs. This can make index lookup expensive even when the final range is selective.

Combine lower and upper bounds on the same field under AND into one bounded BTree scan. Forward the combined range through the union and row-ID offset readers, reuse existing index-file pruning, and preserve inclusive/exclusive endpoints. Other index readers retain the default evaluation of supported bounds.

Tests

• Verify one bounded query on the real BTree reader, with no separate lower/upper queries; reversed ranges are pruned without querying the BTree reader.
• Cover 12 data types, all endpoint combinations, equal/reversed bounds, both predicate orders, null exclusion, and row-ID offsets through union/offset wrappers.
• Verify partial-support fallback and that predicates across OR or different fields are not fused; run existing BTree and bitmap table regression tests.