Bounded Rust SDK batch insert plan

Goal: amortize metadata commits across a finite group of episode rows without routing large object bytes through SQL or Flight.

Task 1: Lock the public contract

1. Add an owned multi-row SDK API and typed empty/limit errors.
2. Keep the existing single-row API source-compatible by delegating to it.
3. Validate every row before any upload begins.

Task 2: Build one bounded Arrow batch

1. Reorder each row into table-schema order after validation.
2. Upload FILE values sequentially through the existing managed store.
3. Build one array per field and one RecordBatch for all rows.
4. Reuse the existing operation identity, digest, retry, and resume path.

Task 3: Prove behavior

1. Commit several distinct files and scalar values in one version.
2. Query every row and directly read every returned DataLocation.
3. Inject a bad later row and prove zero uploads and no version advance.
4. Reject empty/excessive batches before any endpoint or store access.

Task 4: Ship

1. Update the managed FILE example, README, and architecture contract.
2. Run spec lifecycle, strict SDK clippy, full gate, independent review, and independent verification before merge.