Tracking Issue to deliver the channel family incrementally

[tisonkun]

Goal

Deliver the channel taxonomy agreed in #167 through small, independently reviewable pull requests instead of one bundled implementation.

Current status for 0.7

• The source tree and public modules now share direct crate-root families; refactor(channel): simplify channel structure #234 removed the intermediate private channel source wrapper without changing public paths.
• The ecosystem benchmark baseline is complete in test(channel): establish ecosystem performance baselines #208.
• Lossless unbounded MPMC broadcast is implemented at asyncband::broadcast::mpmc::unbounded by feat(broadcast): add unbounded MPMC channel #117; feat(broadcast): add a bounded lossless MPMC channel #213 now tracks only the bounded variant.
• The latest-state watch channel is complete in feat(watch): add a latest-state channel #215 and feat(watch): add a latest-state channel #233.
• The Cargo feature decision is complete: oneshot, mpsc, broadcast, and watch remain independent opt-in features, with no channel umbrella for 0.7. See build(channel): consolidate feature flags and documentation #216 and docs(channel): clarify feature boundaries #226.

Working agreements

• Keep public modules and their source families at direct crate-root paths such as asyncband::mpsc and asyncband/src/mpsc.
• Put machinery shared by multiple families under the private crate-level internal module; keep family-specific machinery inside that family.
• Preserve existing public behavior unless a subtask explicitly proposes an API change.
• Add one channel family or one implementation concern at a time.
• Compare contention and topology-specific performance with semantically similar Rust channels. Being the fastest is not required, but sustained order-of-magnitude regressions must be treated as implementation defects.
• Keep bounded backpressure and unbounded growth as the initial policies. Sliding, drop-oldest, and lag semantics remain deferred.
• Do not expose queue, ring, Disruptor, or sequencer machinery through the public API.

Remaining delivery

• refactor(mpsc): replace the legacy queue backend #209: reconsider the existing MPSC backend.
• design(spsc): decide whether to expose a dedicated queue #210: decide whether an SPSC queue is justified.
• feat(mpmc): add a competing queue #211: add a competing MPMC queue.
• feat(spmc): add a competing queue #212: add a competing SPMC queue after the MPMC design.
• feat(broadcast): add a bounded lossless MPMC channel #213: add bounded lossless MPMC broadcast.
• feat(broadcast): add an SPMC specialization #214: evaluate an SPMC broadcast specialization.

Design baseline: #167. Prior broadcast work: #117 and #146.