refactor: validate scalar Iceberg results with native values

[alexanderbianchi]

Native scalar validation, with less test code

Stacked on #705 (iceberg/test-statistics-cases, base f8dcd0c18f2420602f7ee62524ff88e667c50519). This covers every scalar query-result test in the current Iceberg suite, plus the empty-result case—not just the initial three-function pilot.

Net −14 lines across the whole PR: 7 files, 91 insertions / 105 deletions, including the native query API and shared assertion helper. The initial pilot was +28 lines; the expanded version is 42 lines smaller. All existing test cases remain.

Before / after

Test / behavior	Before	After
Explicit metadata override	ASCII count table	One Int64(0) row, alias trips
Explicit snapshot selection	ASCII count table	One Int64(175000) row, alias trips
Metadata-only COUNT(*), both statistics settings	Plan and result snapshots	Native ProjectionExec -> PlaceholderRowExec checks and Int64(175000) result
Compound predicate pushdown	Plan and count-table snapshots	Same plan snapshot, typed count 9891
Null predicate pushdown	Plan and count-table snapshots	Same plan snapshot, typed count 8829
Unsupported residual filter after safe pushdown	Plan and count-table snapshots	Same plan snapshot, typed count 10516
Wholly unsupported filter	Plan and count-table snapshots	Same plan snapshot, typed count 2757
Computed MAX(trip_distance * fare_amount)	Plan and scalar-table snapshots	Same plan snapshot, Float64(207508.9584) result
LIMIT 0	EmptyExec and empty-table snapshots	Native EmptyExec check and zero total result rows

Eight scalar-result functions (nine named cases) share the same assertion helper. The empty-result case makes ten migrated cases overall. Multi-row/multi-column results, DESCRIBE, errors, statistics, and meaningful plan snapshots remain intact; they are not replaced merely to remove snapshot macros.

Structure

• IcebergTestHarness::query_raw returns the executed native plan and Arrow batches, using one physical plan and its DataFrame task context.
• Existing query() -> Result<(String, String)> stays compatible. It captures plan text before executing that same plan, preserving existing snapshots without resetting runtime state or creating a second physical plan.
• The shared integration-only scalar assertion checks one total row across batches, one column, its alias, and the expected typed value. It does not assume one physical batch. #[track_caller] points failures at the calling test.
• Typed i64/f64 expectations distinguish values that would print identically as table text. No public assertion API, dependencies, feature flags, benchmarks, or root comparator changes. No performance claim.

Validation

• cargo test -p datafusion-distributed-iceberg --locked — 84 tests and 1 doctest passed; no cases removed
• cargo clippy -p datafusion-distributed-iceberg --tests --locked -- -D warnings — passed
• cargo fmt --all -- --check and git diff --check — passed
• Negative value check: expecting Int64(1) for the empty table fails against actual Int64(0)
• Negative type check: expecting Int32(9891) fails against actual Int64(9891) at the filter test's call site, even though both values print as 9891
• Both negative controls were reverted; no snapshot expectations were updated

Validation is scoped to the Iceberg crate; distributed integration/gRPC features were not enabled. #709 remains the independent result-comparator bug fix.