[FEAT] Core log macros reach defmt: forward log_*! to defmt and drop the hand-written mirrors #352

Is your feature request related to a problem? Please describe.

On an Embassy device, core's own log lines never reach the probe log. aimdb-core/src/log.rs forwards the log_*! macros to tracing and log only. defmt was left out on purpose, because many call sites use {:?} on types that don't implement defmt::Format. Places that matter on a device mirror by hand instead (#[cfg(feature = "defmt")] defmt::warn!(…) next to the log_warn!).

That has two costs:

• Events are invisible on the device. The session engine has no mirrors at all. While testing "[BUG] AimX over serial: every other reconnect gets no events until the board is reset" #328 and [BUG] session::serve spins on a listener whose accept keeps failing, and starves every open connection #351 on a NUCLEO-H563ZI, the session reset ("new handshake … ending N subscription(s)"), refused and ended subscriptions, and serve's accept-failure summary ("accept failed …", "still failing: N attempts") could only be seen by their effect.
• Every message exists twice. There are 30 hand-written defmt::…! mirrors (21 in aimdb-mqtt-connector, 5 in aimdb-embassy-adapter, 4 in aimdb-core). They drift, and new log lines get no mirror. That's how the session engine ended up without one.

Describe the solution you'd like

log_trace! … log_error! also forward to defmt under core's defmt feature, as a third destination next to tracing and log. No new logging layer and no change to call sites' format strings.

A check against defmt 1.0.1 (thumbv7em, cargo check) shows that defmt accepts the existing syntax, including {:?} as a display hint, and with defmt's alloc feature it formats String and Vec<String>. So the work is in the argument types, not the call sites.

Scope (measured by parsing every macro call)

Plan: a short PR series, each reviewed separately.

1. defmt::Format for logged argument types, under the defmt feature:
   • cfg_attr(feature = "defmt", derive(defmt::Format)) on TransportError, CodecError, AuthError, RpcError, the framer error in session/io.rs, and the error types logged by the MQTT and KNX connectors;
   • DbError gets a hand-written impl. Its std-only variants carry io::Error / serde_json::Error, which don't implement Format, and docs.rs builds with all-features = true (Cargo.toml:186), so std and defmt are on together there and a derive would break that build;
   • core's defmt feature enables defmt/alloc.
2. The macro branch, plus CI: #[cfg(feature = "defmt")] defmt::…! in each log_*! macro, with the log.rs module docs updated. Direct thumbv7em checks with defmt for aimdb-core (alloc,connector-session,defmt), aimdb-mqtt-connector and aimdb-knx-connector. Today core with defmt is only compiled through the serial connector, so a missing Format impl would surface late.
3. Remove the hand-written mirrors that the macros now cover (MQTT connector, Embassy adapter, core), keeping only those that log something the log_*! line doesn't.

Contract change: the macros are #[macro_export]ed. With defmt on, their arguments must implement defmt::Format. Normal (tracing/log) builds are unaffected. This needs a changelog entry.

On the device:

• Format strings are interned, so they cost almost no flash.
• debug/trace lines are filtered at compile time via DEFMT_LOG, so they cost nothing at the default info level.
• Firmware built with DEFMT_LOG=debug will show about 60 more core lines than today.

Verification

• The new thumbv7em checks.
• Live on a NUCLEO-H563ZI: two aimdb watch clients one after the other over serial show the session reset; a watch on a record without a free subscriber slot shows the ended subscription; a TCP server on port 0 shows serve's accept-failure summary ([BUG] session::serve spins on a listener whose accept keeps failing, and starves every open connection #351).

Describe alternatives you've considered

• Hand-written mirrors in the session engine only (the first plan for this issue): small, but it adds to the duplication above and leaves every other core log line invisible on devices.
• Typed events (SessionEvent, …) instead of text: valuable beyond logging (tests, metrics, introspection), but a separate design question, tracked in its own issue. It doesn't depend on this one.

Additional context

• Builds on [BUG] session::serve spins on a listener whose accept keeps failing, and starves every open connection #351 (serve's accept-failure logging, PR fix(session): a failing listener no longer stalls serve (#351) #356). Start after fix(session): a failing listener no longer stalls serve (#351) #356 is merged.
• Related: "[BUG] AimX over serial: every other reconnect gets no events until the board is reset" #328.

🤖 Generated with Claude Code