refactor: build manifest statistics fixtures in memory

[alexanderbianchi]

Superseded by #705, which combines the test consolidation and populated-manifest coverage in one PR based on #700. The combined version replaces weak checks instead of adding a parallel test suite, includes missing-metric cases, and has a net LoC reduction. See #705 for the before/after coverage grid.

Original proposal

Draft example of #700's generic with_file(uri, bytes) support. Adds two integration tests using manifests and a manifest list serialized by Iceberg's native writers into memory, without new harness methods or dependencies.

The fixture contains two data files with known metrics for vendor_id (Iceberg field ID 1) and passenger_count (ID 4). Tests assert:

• null-count and byte-size aggregation across files;
• merged min/max bounds and their precision;
• preservation of metrics through reordered projection, using non-contiguous field IDs.

Unlike the checked-in taxi manifest, whose column metric maps are empty, this fixture produces populated column statistics. The tests use synthetic data-file paths and only plan queries; they do not claim to validate statistics against actual Parquet contents. File row counts and sizes sum to the existing snapshot summary totals.

Stacked on #700 (codex/iceberg-runtime-metadata). Missing-metric combinations, multiple manifests, delete files, and NDV metadata were outside this initial example.

Original validation

• Both new tests pass with column statistics enabled.
• Temporarily disabling column statistics makes both fail with Absent values; the mutation was reverted.
• cargo test -p datafusion-distributed-iceberg --locked — 92 tests passed, including the doctest
• cargo clippy -p datafusion-distributed-iceberg --tests --locked -- -D warnings
• cargo fmt --all -- --check
• git diff --check