test(channel): add ecosystem performance baselines

[tisonkun]

Summary

• add a standalone ecosystem benchmark target for the current bounded and unbounded MPSC APIs
• compare ready round trips and 1/2/4/8-producer contention with Tokio, async-channel, and flume
• organize Asyncband self-benchmarks and ecosystem comparisons as sibling suites, each grouped by primitive and operation
• keep all peer implementations confined to benchmark dev dependencies

Closes #208.

Design Notes

This PR was stacked on #217; GitHub retargeted it to main after #217 merged.

The benchmark crate root contains only Cargo.toml. Asyncband self-benchmarks live under benchmarks/asyncband/<primitive>/<operation>.rs, while cross-project comparisons live under benchmarks/ecosystem/<primitive>/<operation>.rs. Shared setup stays in the nearest support.rs, and the existing self-benchmark scenarios remain unchanged.

Tokio is the closest MPSC peer. async-channel and flume expose MPMC receivers, so the suite measures them with one receiver and records that topology difference. The compared versions are Tokio 1.53.1, async-channel 2.5.0, and flume 0.12.0; Cargo.lock records the exact resolution.

Bounded channels use capacity 64. Concurrent cases move 16,384 usize messages from 1, 2, 4, or 8 producer threads to one consumer. Channel construction and worker lifecycle stay outside the timed section, and async paths are driven without installing a peer-specific runtime. Run the full workflow with cargo x bench, or focus this target with cargo bench -p benchmarks --bench ecosystem.

These comparisons are regression signals rather than a fastest-wins contest. A sustained result above 3x the closest semantic peer warrants investigation; an order-of-magnitude gap should block the implementation unless a documented semantic or resource tradeoff explains it.