test: benchmark MPMC on Tokio and replace Pollster

[tisonkun]

Summary

MPMC comparisons currently measure dedicated OS threads driving async operations through Pollster. Add Tokio task benchmarks for asyncband, async-channel, and Flume, covering bounded/unbounded queues, current-thread/four-worker runtimes, and four producer/consumer topologies. Share the fixtures between the asyncband-only and ecosystem suites.

Replace Pollster in integration tests and channel benchmarks with the existing asyncband::blocking::FutureExt API, and remove the workspace dependency.

Design Notes

Keep the dedicated-thread workload as blocking_threads: it measures synchronous callers bridging into async channel APIs, including polling and parking overhead. The new tokio_tasks workload runs every producer and consumer in a spawned task; consumers compete until disconnection and drain all messages.

Regression coverage checks completion, received counts/checksums, capacity-one backpressure, uneven consumer counts, and last-sender drop. Related to #298.

Validation

• cargo x lint, cargo x check, cargo x test, cargo +1.86.0 x test, cargo x miri, and cargo x bench --no-run.
• Executed all 72 ecosystem and 24 asyncband-only MPMC cases with 20 samples each, plus smoke checks of the affected MPSC workloads.