feat: add kai.scheduler/v1 group, usagedb structs, podgrouper constants

[SiorMeir]

What

Seeds the api module with the remaining API surface consumed by external users (run:ai), so runai can eventually depend on github.com/kai-scheduler/api alone.

Adds:

• kai/v1 — the full kai.scheduler/v1 group: Config, SchedulingShard + component sub-packages (admission, binder, common, node_scale_adjuster, numa_placement_exporter, pod_group_controller, pod_grouper, prometheus, queue_controller, scheduler), with deepcopy + the two CRDs.
• usagedb — usage-DB config structs (UsageDBConfig, UsageParams, WindowType), split from the scheduler-runtime Interface (which stays in KAI-Scheduler) so this package has no pkg/scheduler dependency.
• podgrouper/constants — PodGroup labeling/annotation constants.

Notes

• CRD manifests kai.scheduler_configs.yaml / kai.scheduler_schedulingshards.yaml are byte-identical to KAI-Scheduler main — zero contract drift.
• No generated client for kai/v1 (controller-runtime only; no +genclient) — make clients is a no-op for it.
• Dependency versions pinned to match KAI-Scheduler (prometheus-operator monitoring v0.88.0, VPA v1.5.1); k8s stays at v0.35.4, controller-runtime v0.23.3 — no transitive bump.
• Additive only; KAI-Scheduler is untouched. Consuming (import repoint) + code removal land as follow-up PRs. Cut as v0.2.0 once merged.

Verification

• go build ./..., make generate manifests clients, make lint, make test — all green.
• CRD byte-diff vs KAI-Scheduler source — identical.
• No kai-scheduler/KAI-scheduler back-imports in the module.