Implement MATCH_RECOGNIZE and pattern time functions

Problem

MATCH_RECOGNIZE still runs in Flink. Its MATCH_ROWTIME() and MATCH_PROCTIME() functions require pattern-match context and cannot be implemented as independent scalar temporal functions.

Scope

Implement the pattern-matching operator as a separate follow-up to temporal scalar coverage. Consult Flink and Arroyo first, document any architectural divergence, and preserve native Arrow batch boundaries.

Admission must explicitly cover supported pattern syntax, partitioning and ordering, measures and definitions, time attributes, skip strategies, timeouts, watermarks, and checkpoint/recovery. Unsupported combinations should remain on Flink.

Validation

• Differential SQL tests against released Flink, including MATCH_ROWTIME and MATCH_PROCTIME context and deterministic arrival-order behavior.
• Stateful and watermark tests, including late input and recovery.
• Release benchmarks with both row/Arrow transposes retained.
• Update operator coverage documentation with the exact admitted shapes and remaining gaps.

Start by consulting Flink and Arroyo to map supported MATCH_RECOGNIZE syntax and identify architectural differences. Define admitted shapes across partitioning, ordering, measures, time attributes, skip strategies, timeouts, watermarks, and recovery, then run the specified differential SQL, stateful, watermark, and benchmark validation. Done means documented coverage, preserved Arrow batch boundaries, and explicit unsupported cases delegated to Flink.