[lxsaah]

Closes #438. Part of #422.

What changes

An exact-topic AimX subscribe over UDS, TCP or serial now starts with the record's current value as a snap frame (last: true), followed by events from seq 2. Wildcard subscribes and WebSocket already work this way. A record with no current value sends no snapshot. That covers records with nothing written yet and SpmcRing buffers.

The duplicate this also fixes

A new SingleLatest reader starts on the record's current value, on tokio (mark_changed) and on embassy (Watch). So:

• exact subscribes already received the current value, but as a plain first event;
• wildcard subscribes received it twice, once as the snapshot and again as the first event.

AimxSession::record_stream now opens the reader and then reads the snapshot, in that order. The stream then drops leading updates stamped at or before the snapshot. A write after the snapshot has a later stamp, so it is still delivered as an event. The snapshot is read with try_latest_as_json_bytes, the same path record.get uses. The engine calls subscribe and then snapshots, so subscribe stores the snapshots and snapshots() hands them over.

Client, CLI, MCP

• watch::arm + await_updates: no change. The snapshot's stamp equals the baseline, so it is filtered out. armed_watch_sees_only_the_write_after_it still passes for SingleLatest, ring and Mailbox.
• aimdb watch <record>: the current value now arrives as a snapshot, followed by the 📸 snapshot complete line. It came as the first event before, so --count counts it as before, and --new still skips it. The README and CHANGELOG are updated.
• MCP: no tool subscribes, so nothing changes.
• I left out the optional simplification of arm (dropping the record.get). It can be a follow-up.

Tests

New test exact_subscribe_sends_the_current_value_once in aimdb-client/tests/aimx_session.rs:

• an exact subscribe gets one snapshot that closes the burst;
• a wildcard matching only that record gets the same snapshot (topic, value and ts);
• after a write, every stream's next update is that write;
• a ring and an unwritten record get no snapshot.

With the head filter turned off, the test fails on the duplicate value.

Locally:

• client, CLI, core (std,tracing), UDS, TCP (_test-tokio) and serial (std) tests;
• clippy on core (alloc,remote,connector-session and std,tracing,observability), client, CLI and MCP;
• the Cortex-M fixture clippy;
• fmt.

Compatibility

Wire change: AimX exact-topic subscribers now receive a snap frame before the first event. In-repo clients already handle snap. The core CHANGELOG entry is marked as a wire change.

🤖 Generated with Claude Code