feature: resolve column statistics across index and seed coverage

Parent: #8461

Goal

Resolve column statistics for a dataset snapshot from valid ZoneMap index coverage and compatible write-time seeds, with explicit unknown results wherever statistics are unavailable. Newly written data should be represented without waiting for central-index consolidation.

Existing foundation

#7463 provides conservative dataset-level min/max from fully covered ZoneMap indices, and #7924 handles bounds invalidated by overlays. #7427 supplies write-time seeds for index maintenance. A shared statistics resolver is still needed to combine these sources for metadata retrieval.

Resolution contract

For each requested column/statistic and relevant fragment:

1. Use valid index statistics for the requested snapshot.
2. Fill uncovered portions from compatible seeds, without double counting index-covered data.
3. Return explicit unknown or incomplete coverage where neither source is usable.

Coverage and precision are separate. Full fragment coverage does not imply exact live-value statistics: deletions may leave conservative extrema and counts that describe earlier physical values. Updates or overlays may invalidate bounds. All-null data, unsupported statistics, missing statistics, and stale statistics must remain distinguishable.

Acceptance criteria

• Resolve mixed index/seed coverage per column and statistic against the requested dataset snapshot.
• Validate field identity, type, seed format, and relevant configuration before using a seed.
• Avoid duplicate contributions when index segments and seeds overlap.
• Handle append, deletion, overlays/updates, rewrites, and schema evolution without reporting invalid statistics as current.
• Expose completeness and precision independently; never present partial aggregates as complete dataset statistics.
• Return unsupported or unavailable statistics explicitly, including null-only and NaN cases.
• Retrieval does not implicitly scan column values, rebuild indices, or mutate the dataset.
• Tests cover multiple fragments and segments, mixed per-column coverage, all-null/NaN values, and mutations that introduce values outside old bounds.

Boundaries

The public typed API is tracked in #4163. Automatic seed collection belongs to #4522, and consolidation/backfill is another child of #8461. This issue supplies statistics resolution; query planning and execution changes are outside its scope. Derived or cached summaries can be evaluated without requiring a new persisted fragment-statistics format.

Start by reading the foundation issues #7463, #7924, and #7427, then compare their coverage, invalidation, and seed behavior with the resolution contract here. Done means mixed index/seed statistics resolve per column and snapshot without duplicate contributions, while completeness, precision, unsupported cases, and mutations remain explicit. The linked boundaries identify #4163, #4522, and #8461 as related but out of scope.