feat(sim): run the receiver firmware and inject RF frame loss

[FredM67]

Stage 2 of the RF mock, following #171. The remote units now run the real RemoteLoadReceiver firmware, and RF frames can be lost on purpose.

Changes

• Radio model, receive side (rfm69.c): a frame that has been sent reaches every other radio.
  • A radio listening (RX mode) on the same network gets the frame in its FIFO, with PayloadReady set and DIO0 raised on D2 (INT0). The library's interrupt and receiveDone() path therefore runs unmodified.
  • A radio that isn't listening misses the frame, and this is counted.
• Real receivers: grid_sim -r <node>=<receiver.elf> runs the receiver firmware in its own simulated AVR, in step with the router on the same clock.
  • Its load pins (receiver_pins 4 3) drive the remote loads.
  • Its serial output is shown, prefixed R<node>|, and its "RF link LOST" messages are counted.
  • Nodes without firmware keep the ideal receiver.
• Frame loss:
  • rf_drop <t0> <t1> [node] loses every frame in a window.
  • rf_loss <percent> loses frames at random, from a fixed seed, so every run loses the same frames.
  • expect rf_links_lost <node> <min> <max> checks how often a receiver lost the link.
• build_receivers.sh: builds the receiver once per node ID, since REMOTE_NODE_ID is a constant in its config_rf.h, and restores the file afterwards.
• Makefile and CI: make check-rf uses the receiver firmware for RX_NODES (15 by default); RX_NODES= switches back to ideal receivers. The CI step is unchanged apart from its name.

Results with the receiver firmware

Scenario	Result
remote_override	139/139 frames received; 5/5
remote_step	744/744 frames received, grid regulated through the remote load; 10/10
link_loss (new): no frames to node 15 from 10 to 11 s	load off at 10.41 s (500 ms after the last frame), back on at 11.00 s, link lost once; 6/6
lossy_link (new): 20 % of frames lost	54 of 239 frames lost, link never lost; 4/4

The ideal receivers give the same results in all four scenarios. No frame was missed because a receiver wasn't listening.

Each receiver roughly halves the simulation speed: check-rf takes ~50 s with one receiver.

Next, on #160

• Set RX_NODES to 15 16.
• Add node 16 to the loss scenarios, for example losing unit 1's frames while unit 2 keeps its load.

🤖 Generated with Claude Code