fix(bench): respect Notice delivery capacity

[smiggleworth]

Summary

• keep each Tier 2 Notice publish batch within the production per-family delivery lane capacity
• allow bounded delivery draining before asserting completed fanout
• preserve the benchmark as a completed-fanout measurement instead of asserting against intentional best-effort drops

Root cause

The benchmark enqueued 512 publish jobs per batch into a production delivery lane with capacity 64. Notice intentionally uses non-blocking delivery, so saturation dropped jobs before the benchmark drain assertion. The scheduled run reproduced 6,432/8,192 and 5,392/8,192 deliveries; raising only the timeout still reproduced 3,936/8,192 locally.

Validation

• cargo fmt --all -- --check
• cntryl-tools validate-benchmarks
• cargo test --workspace (1,612 library tests plus all integration and doc suites)
• cargo clippy --locked --workspace --all-targets --all-features -- -D warnings -D clippy::pedantic
• formerly failing double-star/16-subscriber workload passed
• complete tier2_subsystem_notice benchmark passed all 9 rows