perf(database): integrate CompactResultDataset into SQL drivers to eliminate 3x memory copies and string allocations #1355

Summary

lib/core/database/compact_result_dataset.dart implements a columnar dataset CompactResultDataset with typed arrays (Int64List, Float64List), bit-packed null bitmaps, string interning, and lazy on-demand string generation via asLazyRowList(). It is fully tested in test/core/database/compact_result_dataset_test.dart, but is currently completely unused in production code.

Instead, PostgreSQL, MySQL, and SQLite drivers eagerly convert all query results into a full List<List<String>> matrix:

1. Driver raw records (Result / taken.items / results).
2. Intermediate rawRows (List<List<Object?>>).
3. Eagerly stringified outRows (List<List<String>>).

For a 10,000–50,000 row result set across 30 columns, this creates up to 1,500,000 individual String objects on the Dart VM heap, taking 150–300 MB of RAM and causing large GC pauses, instead of 10–20 MB when packed in columnar typed data.

Proposed Solution

1. Integrate CompactResultDataset into SqlExecutionResult, or construct CompactResultDataset.fromRawRows directly in PostgresSqlExecutionDelegate, MySqlSqlExecutionDelegate, and SqliteSqlExecutionDelegate.
2. Provide rows to VirtualResultGrid via dataset.asLazyRowList(), so only visible cells in the viewport are stringified.

Acceptance Criteria

• Memory footprint for queries returning $\ge 10,000$ rows is reduced by $\ge 60%$.
• Intermediate List<List<Object?>> and List<List<String>> allocations are eliminated in favor of columnar typed storage.