Release prep v0.7.2: relay_node tracking (PR #45) + CRC_BAD packet drop (closes #34)

[KMX415]

Summary

Prep branch for the v0.7.2 release. Bundles two changes that have been ready to ship since May 1:

1. Relay-node debugging on the dashboard (originally PR Added information for debugging mesh connects #45, accidentally closed by the contributor on May 4)
2. Drop CRC_BAD packets at source (the fix/crc-bad-drop branch, never opened as a PR)

Version bump and CHANGELOG entry are deliberately not in this branch yet — they will land as a final commit once hardware validation on the test RAK is green. That keeps the fleet-wide update indicator (__version__ on main) from firing until we've confirmed the bundle runs cleanly on real silicon.

What's in this branch

89b3dbe  KMX415          Drop CRC_BAD packets at source instead of passing them to decoder
1c19534  KMX415          Add relay_node tests and small review polish
2c77d91  kendelmccarley  Add relay_node tracking and packet-focus map line  (Co-Authored-By: Claude Sonnet 4.6)
1627b3f  kendelmccarley  Fix: read hop_limit from TransmitConfig instead of hardcoding 3  (Co-Authored-By: Claude Sonnet 4.6)
─── 1779b66  Fix coding rate on Short and Medium presets, add LongTurbo (#44)

Co-authored-by: trailers preserved per the May 4 attribution policy ("contributors choose their own attribution").

1. Relay-node tracking (originally #45)

Surfaces the Meshtastic header relay_node byte (the lowest byte of the last relay node's ID) through the decoder → Packet model → SQLite schema (with ALTER TABLE migration for existing installs) → WebSocket payload → dashboard packet feed. Clicking a row in the packet feed now draws a line on the map between the source and the relay, and the source cell shows !src ↝ !relay with full short-ID resolution when the relay is in the local node registry.

Real-world utility: kendel used this to trace a hop chain back to a rooftop node and discovered its ERP was bad.

Bonus fix in the same series: text-message and NodeInfo TX paths now read hop_limit from TransmitConfig instead of hardcoding 3 in two places. Behavior unchanged for anyone running the default (still 3); fixes the silent "I set hop_limit in my yaml and it didn't take" bug.

2. CRC_BAD packets dropped at the HAL boundary

SX1302Wrapper.receive() was logging the warning for STAT_CRC_BAD packets but still appending them to the returned packet list. The crc_ok=False field on ConcentratorPacket has never been read by any downstream code (concentrator_source, packet_router, decoders), so RF-corrupted bytes flowed all the way into MeshtasticDecoder.decode().

Three observable downstream symptoms produced by this:

1. Phantom node IDs in the local SQLite node table and the cloud DynamoDB node catalog. A bit flip in the source-ID field creates a "new node" that shares all-but-one bit with a real source ID.
2. False ENCRYPTED packets. A corrupted channel-hash byte stops matching the LongFast hash and the packet is filed under "encrypted on a private channel we lack the key for."
3. Garbled-but-readable text. AES-CTR XORs corrupted ciphertext with the keystream and produces mostly-correct plaintext with a few mangled characters. This is the "3 hopr frOO\"Nc >> Mesa" symptom in Parsing Error on Some messages #34.

This bug predates v0.7.0 (the pre-source-publication HAL didn't even define STAT_CRC_BAD), but v0.7.0's RX diagnostic logging made it newly visible. The append-anyway behavior has been live since the initial RAK2287 wiring commit.

Fix is one line: continue after the CRC_BAD log. The diagnostic WARNING and crc_bad_count counter both keep working unchanged. STAT_CRC_OK and STAT_NO_CRC paths are deliberately untouched.

Closes #34.

Cloud-side evidence (informational)

A diagnostic audit on meshradar.io this morning found the cloud active_nodes_24h = 18,374 count is dominated by phantom IDs that match the CRC_BAD-passthrough fingerprint:

• 89.6% of "active" nodes were heard by exactly one Meshpoint (real RF mesh nodes overlap across antennas)
• 76.7% have no long_name AND no position (decryption fails on corrupted ciphertext → ENCRYPTED with no decoded payload)
• 24.2% at hop_count ≥ 4 (the hop_count is 3 bits in the flags byte — a single bit flip promotes hop=0 → hop=4)
• Top contributors are the busiest-RF Meshpoints in the fleet

After this fix lands and the fleet rolls forward, expect active_nodes_24h to drop sharply (rough estimate: from 18k toward 5–8k) over the following 24–48h as phantom IDs age out of the 24h window. We'll watch the trend post-deploy and decide whether to run a one-shot purge of pre-fix phantoms in DynamoDB or just let the 30-day NODE_TTL clean them up naturally.

Test plan

• pytest tests/ → 297 passed, 1 pre-existing warning, 2.60s
• ruff check . → all checks passed
• Hardware validation on RAK V2 .141 from this branch:
  • git checkout release/v0.7.2, sudo systemctl restart meshpoint, watch logs for clean startup
  • MESHPOINT_DEBUG_RX=1 for 5 minutes — confirm STAT_CRC_BAD lines no longer carry an associated phantom-node DB write
  • Send Meshtastic DM to/from .141 and validate relay_node populates correctly in the dashboard packet feed
  • Click a packet row, confirm the map line draws between source and relay
  • Verify meshpoint logs for any new tracebacks
• After hardware-green: append a single Release v0.7.2 commit bumping __version__, default.yaml firmware_version, README badge, and CHANGELOG. Push, watch CI, squash-merge.

Authoring + attribution

Kendel's two commits on this branch (1627b3f and 2c77d91) carry his GitHub noreply email so his profile links correctly in the contributors graph + git blame, identical to the PR #38 squash on main. The Co-Authored-By: Claude Sonnet 4.6 trailers from his original commits are preserved.

PR #45 was accidentally closed by the contributor on May 4 with a misdirected force-push that also wiped the work from his fork. Local pr-45-kendelmccarley retained everything, and these commits are cherry-picked from there. A linkback comment will be posted on the closed PR pointing here so the original contribution thread stays connected.