Backplane stops every board at startup when all transports close (e.g. MUJINA_USB_DISABLE=1 with the CPU miner)

[Jolah1]

Issue Description

With USB discovery disabled and the CPU miner enabled (the README Quick Start settings), the backplane shuts the CPU board down a few milliseconds after starting it.

This is a regression from cd96476 (feat: hold startup until initial board enumeration completes). Since then each transport has its own channel. The CPU transport sends its device and InitialEnumerationComplete, then drops
cpu_tx (daemon.rs). When it is the only transport, the backplane's StreamMap ends, Backplane::run() returns Ok(()), and the spawned task falls through to shutdown_all_boards(), which is meant to run only on shutdown. Before
cd96476, transport_tx lived for the whole of Daemon::run, so the stream never ended. I confirmed this by building cd96476^ (board stops only on SIGINT) and cd96476 (board stops at startup).

Impact:

• Today the CPU board has shutdown: None, so its threads keep hashing. But the backplane forgets the board, so at exit it never runs the board's shutdown. That's harmless today only because the CPU board has no shutdown work.
• PR cpu-miner: make the CPU backend a first-class citizen for API/tooling #56 gives the CPU board a real shutdown hook. With cpu-miner: make the CPU backend a first-class citizen for API/tooling #56's board changes merged onto current main, /api/v0/boards is still [] in Quick Start mode, because the hook fires at startup. With USB enabled, the same build lists the
  board. cpu-miner: make the CPU backend a first-class citizen for API/tooling #56 was tested in May, before cd96476.
• Any future board on a transport that closes after enumeration would be shut down the same way.

I have a fix ready: keep boards running until shutdown is requested, even after every transport closes. It comes with a regression test, following the should_panic pattern in CONTRIBUTING.md. I'm happy to open a PR if this
sounds right.

Expected Behavior

Boards keep running until the daemon receives a shutdown signal. "Board stopped" is logged only on shutdown, as it is when USB discovery is enabled.

Actual Behavior

"Board stopped" is logged in the same millisecond as "Board started". On SIGINT, no "Board stopped" is logged, because the backplane has already forgotten the board.

Reproduction Steps

1. On main (9801285), run:
   MUJINA_CPUMINER_THREADS=1 MUJINA_CPUMINER_DUTY=50 MUJINA_USB_DISABLE=1 cargo run --bin mujina-minerd
2. Observe "Board started." immediately followed by "Board stopped".
3. Press Ctrl-C: no "Board stopped" at exit.
4. For comparison, rerun without MUJINA_USB_DISABLE: "Board stopped" appears only after Ctrl-C.

Logs

13:48:12 INFO  backplane: CPU miner board connected.  board=CPU Miner, threads=1, duty=50
  13:48:12 INFO  daemon: Started.
  13:48:12 INFO  backplane: Board started.   board=CPU Miner, serial=cpu-1x50%, threads=1
  13:48:12 INFO  backplane: Board stopped    board=CPU Miner, serial=cpu-1x50%
  13:48:12 INFO  api::server: API server listening.  url=http://127.0.0.1:7785
  ...
  13:49:45 INFO  daemon: Received SIGINT.
  13:49:45 INFO  daemon: Exiting.

mujina-miner Version

cargo 1.99.0 (5f94df478 2026-08-27)
git commit: 9801285

Linux Distribution and Kernel

Ubuntu 26.04.1 LTS, Linux 7.0.0-38-generic x86_64

Hardware Info

None. CPU miner only, no hashboard attached.

Configuration

MUJINA_CPUMINER_THREADS=1
  MUJINA_CPUMINER_DUTY=50
  MUJINA_USB_DISABLE=1