perf: benchmark capture-heavy consequence matching

[luantaraschi]

Problem

Progress consequences currently inspect pending obligations in a partition when capture bindings differ. We have evidence for deadline insertion scaling, but not for application-shaped capture matching.

Evidence requested

• A deterministic schema with several pending obligations in one partition.
• Recorded environment and methodology.
• Results across increasing obligation counts.
• A profile showing where time is spent.

Guardrails

Do not add an index until the benchmark demonstrates a real scaling problem. Any optimization must preserve differential finite/online semantics.