Fix Unicode pruning and update dependencies

[platypii]

Unicode range filters could omit matching rows because Iceberg's UTF-8 bounds and SQL's UTF-16 comparisons order some characters differently. Use query ordering for exact partition comparisons and retain uncertain string ranges during metadata pruning. Regression tests compare rows and counts with in-memory SQL across unpartitioned, identity, and truncate partitions, before and after manifest rewrites.

Update hyparquet to 1.31.3, hyparquet-writer to 0.16.11 (fixing Unicode string statistics), AWS credential providers to 3.1147.0, and eslint-plugin-jsdoc to 65.2.0.

Validation: all 947 tests pass, including mixed-Unicode equality and range queries; lint, TypeScript checking, declaration generation, and coverage pass. npm outdated reports no outdated dependencies.