docs(design): revise 054 against the tree after 055

[lxsaah]

Description

Revises design 054 (zero-allocation connector boundary). The doc was written before 055 landed, and a few of its claims about the MQTT connector didn't match the code. This PR changes only the doc, no code.

Rebased on 055

• InboundDispatch::new now takes a &'static dyn TopicGrammar and returns DbResult, matching AimDb::inbound_router. subscriptions() replaces resource_ids().
• The four 055 bench rows (inbound_route_pattern, _keyed_known, _keyed_new) are added to the baseline and target tables. G1 now says "steady state" and excludes the first sighting of a key.

MQTT connector: corrected against the code

• Inbound copies the topic on both backends: embedded topic_name.to_string(), native publish.topic.clone(). The native clone could simply be a move and is called out as a standalone fix.
• The embedded outbound path allocates a Vec<u8> per packet in session_loop::encode for the Channel<Vec<u8>, 4> write queue. The byte ring in §4.6 didn't remove that, so "allocation-free MQTT path" was false. The embedded design now encodes PUBLISH packets straight into a write ring that replaces the channel.
• Native outbound lists two options: ReusableBoxFuture (the default, keeps backpressure) or try_publish (errors reported immediately, no readiness signal).

RoutePublisher semantics

• Adds poll_flush, a flush on shutdown, and explicit error attribution for asynchronous hand-over (an error surfaces one message late).
• The pump loop now waits for poll_ready before receiving, the usual Sink order.

Other

• Drops the async recv/recv_into instead of promising an adapter. The only implementors are in this repo.
• G3: CI only builds b0_alloc_connector today and never runs it. A CI job that runs it is added in migration step 1.
• Notes that the embedded EventChannel stops absorbing bursts once ingest calls dispatch directly.
• Adds Zenoh (053) to the migration plan, a new alternative (a heapless action channel), and new open questions (reserved ring space for control packets, packet-id reservation).

Related Issue

• Follow-up to docs(design): 054 zero-alloc connector boundary, 055 wildcard inbound links, 056 aimdb join CLI #265 / feat: wildcard inbound links (055) #267

Checklist

• I have read the CONTRIBUTING.md document.
• My code follows the project's coding standards. (docs only)
• I have added tests to cover my changes. (n/a, docs only)
• All new and existing tests passed (make check). (not run; no code changes)
• I have updated the documentation accordingly.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RADQu7JwhKvSpicW4qzFz8

---

Generated by Claude Code