IMEX mock on the NRI foundation (channels, daemon, ComputeDomain demo)

[ArangoGutierrez]

Part of the NRI node-wide injection epic (v0.3.0 flagship), building on PR #433. The main payoff of the NRI direction.

Real IMEX (GB200/NVL72-class multi-node NVLink domains) involves per-node nvidia-imex daemons coordinating across the domain, IMEX channel device nodes (/dev/nvidia-caps-imex-channels/channelN) delivered into containers via the DRA driver's ComputeDomain (CDI), and NVML fabric APIs. Without hardware, the hard part is getting channel devices into arbitrary containers — which the NRI plugin now solves generically.

Scope:

• Inject mock /dev/nvidia-caps-imex-channels/* device nodes per container via the NRI plugin (annotation- or claim-driven).
• Mock nvidia-imex daemon reusing the mockib daemon pattern (per-node daemon + Unix socket + cross-node coordination via Service), including nvidia-imex-ctl status surface.
• Multi-node ComputeDomain-style demo on kind (GB200/GB300 profiles; builds on the fabricmanager/NVSwitch NVML surface from feat(nvml-mock): simulate NVSwitch topology and nvidia-fabricmanager #387).

Acceptance: a workload spanning two kind nodes sees consistent mock IMEX channels and a healthy mock IMEX domain status.