Rename skyhook -> nodewright in operator event, log, and error text

[ayuskauskas]

Part of #223. Follow-up from #402 (review comment).

#349 renamed the API group to nodewright.nvidia.com, but the operator's user-visible text still says "skyhook" in a lot of places. #402 fixed the one event string it added and deliberately left the rest, because sweeping them would collide with the other in-flight branches in this epic. This is the cleanup issue for that.

Important

Do this after the jobs epic merges, not during. Almost every site below sits in skyhook_controller.go, job_controller.go or pod_controller.go — the three files every open branch in #223 is editing. A sweep now guarantees conflicts for no functional gain.

Scope: strings a user reads, not identifiers

Counts from operator/ (excluding _test.go):

Surface	Count	Example
Event messages	6	"Package [%s:%s] state %s on [skyhook:%s]"
fmt.Errorf text	33	"error validating skyhook [%s]: %w"
Log messages / keys	11	logger.Error(err, "error getting skyhooks"), "skyhook", name

The 6 event strings are the sharpest, since they land on Node objects and show up in kubectl describe node:

• pod_controller.go:177
• job_controller.go:331
• skyhook_controller.go:2175, :2205, :2296, :3039

(job_controller.go:543 is already [nodewright:%s] as of #402 — the odd one out until this lands.)

Log keys matter as much as messages: logger.Error(err, "error processing skyhook", "skyhook", name) puts skyhook= in structured output that operators may already be querying on. Renaming the key is a small observability break worth calling out in the changelog, so decide deliberately rather than sweeping it in silently.

Explicitly out of scope

• Go identifiers — SkyhookReconciler, skyhookNode, SkyhookOperatorOptions, wrapper.SkyhookNode, EventsReasonSkyhook*, filenames like skyhook_controller.go. That's a mechanical but enormous refactor and belongs in its own issue, if at all. This issue is text only.
• Agent contract surfaces — SKYHOOK_RESOURCE_ID, SKYHOOK_LOG_DIR, SKYHOOK_ROOT_DIR, /skyhook-package/* mount paths, COPY_DIR_ROOT / AGENT_LOG_ROOT defaults (/var/lib/skyhook, /var/log/skyhook). Renaming these breaks running packages and host state; per .claude/CLAUDE.md the agent hasn't migrated yet. Leave them.
• The legacy migration shims — zz.migration.*.go, migration_hold.go. Their text says "Skyhook" because it is genuinely talking about the legacy skyhook.nvidia.com resources. Correct as-is.
• The CLI (internal/cli/, cmd/cli/, 8 more Errorf sites). It ships and versions separately and kubectl skyhook is still the command name; renaming its text is a separate call tied to whenever the plugin is renamed.

Suggested split

1. The 6 event messages — smallest, highest visibility, easiest to review.
2. fmt.Errorf text.
3. Log messages, plus an explicit decision on the "skyhook" structured key.

Done when

• No literal skyhook remains in operator event/error/log text outside the exclusions above.
• The structured-log key decision is recorded in operator/CHANGELOG.md if it changes.
• make unit-tests passes — several tests assert on message text (e.g. k8s-tests/chainsaw/** event assertions), so grep those too.

[mohityadav8]

/assign