[A13xB0]

For #774. Stacked on #788; retarget to main once that merges.

With MESHBENCH_RENODE_LOCKSTEP=1, an nRF52 board under Renode stops running free on the host clock:

• A clock entry in the machine's own virtual time calls the peripheral once a millisecond. It ticks the chip there, then holds the CPU until the engine has granted the next millisecond.
• Advance only grants time and waits for the CPU to reach it.
• emulation SetAdvanceImmediately true, so idle time is skipped rather than waited out.
• emulation SetSeed, derived from the node, because the nRF52 RNG peripheral (MeshCore's jitter source) otherwise draws a fresh stream every run.

This needs no change to the Renode fork, only our peripheral and script.

Measured. Two RAK_4631 repeaters, adverts placed with at_ms:

• Two idle runs gave byte-identical event logs and identical irq_reads/busy_reads per node.
• The run before the seed was added differed, which is how the RNG was found.

Why it stays opt-in, and sim.state still says not reproducible. Console input reaches Renode's UART over a socket on the wall clock. So boot provisioning and console.type land wherever the machine is. A run with all 12 cores busy diverged for exactly that reason; the console port did not even open in time. Lockstep also costs about 5× real time for two boards here.

Remaining on #774:

• Gate the console by virtual time.
• The QEMU half: -icount, plus a hold in the fork's SX1262 device.
• Then drop the emulator case from NotReproducible, with the idle-versus-loaded check.