Retire the legacy NodeInfo, INodeStore, and IContactStore directory model. Route contact projections, protocol observations, persistence, and platform composition through IMeshPeerDirectory across ESP, Linux, and nRF52.
Split LXMF runtime ownership into focused peer, path, link, propagation, and delivery helpers while keeping the adapter as the coordination layer.
Make chat delivery refs and send-result events protocol-aware across Reticulum, Meshtastic, and MeshCore, including Sent/Delivered/Failed semantics, UI badges, retry lookup, and store persistence.
Fix Meshtastic MQTT private replies to publish encrypted direct packets and cover the new delivery behavior with smoke tests.
Add Reticulum token crypto support and conformance coverage, improve LXMF destination/link handling, expand contact/chat UI behavior, and rework Network/Nomad page UI with a Micron renderer.
Keep the Network renderer RAM-conscious by moving page link contexts to a bounded PSRAM-preferred pool and reducing Micron text/table object churn.
Add protocol-scoped MQTT/channel settings for Meshtastic and MeshCore, centralize Settings identity/visibility through SettingId specs, remove BLE from the ESP Arduino firmware profile, and keep Settings Wi-Fi scan/channel/backup paths on fixed buffers.
Centralize Wi-Fi/HTTP ownership for MQTT, Reticulum gateway, firmware updates, route assets, and pack downloads.
Expose Reticulum display name from the configured user name and show it in Settings.
Add Reticulum as a first-class product protocol with LXMF delivery over LoRa and Wi-Fi TCP gateway carriers, SD-backed groups, identity/settings UI, Team appdata support, and conformance guardrails.
Remove the user-facing PC Link surface and keep RNode/LXMF as internal Reticulum implementation details.
Tune Arduino task stacks and board env flags to recover internal RAM for BLE startup.
Avoid synchronous boot UI rendering on Arduino loopTask and move large board singletons toward PSRAM where available.
Improve TDeck/Pager display responsiveness by restoring practical LVGL DMA buffers, using SPI pixel writes, disabling transition animations, and testing TDeck display SPI at 80MHz.
liu weikai