Switch Zephyr nightly workflow to self-hosted runner

[quic-areg]

No description provided.

[Jonathon Penix (jonathonpenix)]

I just went through the exercise of turning these tests on for CPULLVM, so thought I'd leave a few drive-by comments in case they are helpful:

• I'm assuming this change is to help with execution time? FWIW in CPULLVM I tried just using Github runners then splitting the tests up by platform + --subset. Doing that, the total runtime for qemu_riscv32/qemu_riscv64/qemu_cortex_m3/qemu_cortex_a53 is like ~2.75 hours, which isn't too bad.
  
  We only have two self hosted runners right now in CPULLVM, so this seems like a decent tradeoff for us to prevent tying up a runner for many hours at a time--not sure what the situation in eld is, but might be worth trying?
• Re: the dependencies, I went ahead and gave the Zephyr Docker images a shot and, while there's a few "gotchas", overall it went pretty smoothly. I'm not entirely sure how containers work with self-hosted runners, but it might be worth considering (you shouldn't have to worry about updating the dependencies as Zephyr changes and ex: the "developer" Docker container already has the Zephyr SDK pre-installed so you shouldn't have to worry about sourcing that, etc.).

[Shakti Suman (quic-shaksuma)]

Would this not run on self-hosted runner?