tech-debt: extract shared LoRa reconnect attempt + RF auto-connect runner

[rinchen]

Context

Follow-up from the 2026-08-06 24h change review on liam (e807892e^..HEAD). Controllers and production auto-connect routing landed; attempt bodies and ConnectionPanel serial/BLE paths remain duplicated.

Status (2026-08-06 audit): still needed — keep open. Goals 1–2 are partial; goal 3 is still needed (upstream PR only).

Goals

1. Shared LoRa reconnect attempt runner — PARTIAL

   • Done: rfReconnectController single-owner schedule / link-lost / begin–end attempt (fix: packaging smoke path, RF reconnect ownership, and CI build-channel stamps #798); parity contract loraRfReconnectParity.contract.test.ts.
   • Still needed: extract the ~250–320 line parallel attemptReconnect / attemptMeshcoreReconnect bodies from useMeshtasticRuntime.ts / useMeshcoreRuntime.ts behind the controller (keep MeshCore TCP/burst MeshCore-only). Migrate off parallel legacy *IsReconnectingRef / *GenerationRef where safe. (~640 lines of mirrored attempt logic remain.)

2. Shared RF auto-connect transport runner — PARTIAL

   • Done: useProtocolRfAutoConnect + ProtocolAutoConnectCoordinator own production auto-connect (feat: TCP/HTTP RF auto-connect and MeshCore SoftAP reconnect hardening #806); App mounts with suppressMountAutoConnect; TCP/HTTP in the hook; cancel-on-manual-connect.
   • Still needed: dedupe or delete the ~260-line serial/BLE mount auto-connect block in ConnectionPanel.tsx (still mirrored for the unsuppressed / test path). Prefer a shared transport-runner lib or drop the panel path once tests target the hook/coordinator.

3. Upstream rsReticulum bond-desync overlay — STILL NEEDED (upstream process)

   • Done locally: overlay rsReticulum-ble-rnode-bond-desync.patch + apply scripts; mesh-client integration (fix: MeshCore SoftAP TX, hop goblins, Reticulum BLE bond desync #808).
   • Still needed: open/track upstream PR on ratspeak/rsReticulum (chains after pairing-transition-debounce / #20). scripts/update.sh RATSPEAK_PATCH_ENTRIES PR field is still empty; patches README: “no upstream PR yet”.

Related (lower priority in same epic)

• Reticulum remediation i18n consolidation (connectionPanel.reticulumSidecarIssues.* vs diagnostics vs local-interface bond-stale copy)
• Extract MeshCore TCP session / dead-bridge policy out of the large MeshCore runtime
• Prefer pure-lib behavioral tests over expanding source-contract substring scans

Risk

High for reconnect extract (n7eal TCP / single-owner invariant). Do as a dedicated PR with existing loraRfReconnectParity.contract.test.ts + runtime reconnect tests green.

Highest-value remaining

1. Extract reconnect attempt bodies behind the controller (high risk).
2. Delete or share ConnectionPanel serial/BLE auto-connect now that App always suppresses it.
3. Open/track upstream rsReticulum bond-desync PR (process, not local code).

[rinchen]

Goal 3 (upstream rsReticulum bond-desync overlay) is now opened and tracked upstream:

• Upstream PR: fix(ble): halt RNode reconnect on bond desync ratspeak/rsReticulum#21 — fix(ble): halt RNode reconnect on bond desync.
• Opened standalone off main (no dependency on Implement a Cursor on Target (CoT) gateway service to add ATAK compatibility #20 — the bond-desync code does not reference PAIRING_TRANSITION_RETRY_WAIT). Builds with --features ble and test_bond_removed_error_detected passes.
• mesh-client tracking updated: RATSPEAK_PATCH_ENTRIES in scripts/update.sh and the Upstream PR field in reticulum-sidecar/patches/README.md now point at fix(ble): halt RNode reconnect on bond desync ratspeak/rsReticulum#21 (previously empty / "no upstream PR yet"). pnpm run update now reports this overlay as "upstream PR still open" instead of untracked.

Keep the local overlay until #21 merges and floated origin/main includes it; then sunset per the patches README.

[rinchen]

fix up in a PR