feat(ci): add modular nightly build workflow for S3 + DockerHub publishing

[sangli-pensando]

Summary

• Adds a nightly build pipeline (nightly-build.yml) that builds all gpu-operator artifacts, publishes to S3 and DockerHub (amdpsdo/), and calls the common-infra-operator orchestrator for validation
• Workflow is decomposed into reusable stages (build-images.yml, build-helm.yml, publish.yml, build-summary.yml) to enable future extraction to common-infra-operator for cross-repo reuse
• Builds 4 container images (operator, utils, OLM bundle, remediation-configmap-util) + 1 Helm chart, with dedup based on date + SHA

Workflow architecture

nightly-build.yml (orchestrator)
  resolve          — version, date, SHA, S3 prefix, dedup check
  build-images.yml — 4 container images inside dev container (parallel)
  build-helm.yml   — Helm chart inside dev container (parallel)
  publish.yml      — S3 upload + DockerHub push + .build-ok marker
  validate         — calls common-infra-operator/rocm-orchestrator.yml
  build-summary.yml — step summary with all stage results

Artifacts published (nightly)

Artifact	S3	DockerHub
gpu-operator	gpu-operator-{VER}-{DATE}.tar.gz	amdpsdo/gpu-operator:{VER}-{DATE}
gpu-operator-utils	gpu-operator-utils-{VER}-{DATE}.tar.gz	amdpsdo/gpu-operator-utils:{VER}-{DATE}
gpu-operator-bundle (OLM)	gpu-operator-olm-bundle-{VER}-{DATE}.tar.gz	amdpsdo/gpu-operator-bundle:{VER}-{DATE}
gpu-operator-remediation-configmap-util	...{VER}-{DATE}.tar.gz	amdpsdo/gpu-operator-remediation-configmap-util:{VER}-{DATE}
gpu-operator-helm-k8s chart	gpu-operator-helm-k8s-{VER}-{DATE}.tgz	N/A
.build-ok marker	written to S3 prefix	N/A

S3 path: s3://THEROCK_BUCKET_NAME/gpu-operator/nightly/{YYYYMMDD}/

PR validation approach

A temporary pull_request trigger (path-filtered to workflow files) has been added to validate the build stages on this PR. This trigger will be removed before merging.

Expected PR run results

Stage	Expected	Reason
Resolve	pass	Reads Makefile, computes date/SHA
Build Images	pass	Builds dev container + 4 images inside it
Build Helm	pass	Builds dev container + Helm chart
Publish	skipped	Skipped on pull_request events (fork PRs lack repo secrets)
Validate	skipped	Depends on publish
Summary	pass	Reports stage results

Full end-to-end validation (post-merge)

After merging, trigger via gh workflow run nightly-build.yml to validate the complete pipeline including S3 upload and DockerHub push.

Test plan

• PR triggers workflow — resolve + build stages pass
• Post-merge: trigger via workflow_dispatch and verify full pipeline
• Verify artifacts appear in S3 under gpu-operator/nightly/{DATE}/
• Verify images are pushed to docker.io/amdpsdo/ with correct tags
• Verify .build-ok marker is written to S3
• Verify dedup skips rebuild on second trigger with same SHA + date
• Remove temporary pull_request trigger before or after merge

Notes

• Does NOT touch existing weekly-build.yml or ci.yml
• No ROCm tarball dependency (gpu-operator is pure Go)
• Dedup: schedule-only, based on {date}-{sha} cache key
• Pre-release trigger and Helm OCI push are out of scope (nightly only)