feat(action): add deploy-helm-preview composite action

[kriscoleman]

Summary

• Adds a new composite action deploy-helm-preview that manages ephemeral Helm preview environments on Replicated CMX clusters for pull requests
• Implements the full deploy lifecycle: sticky PR comment → GitHub deployment creation → cluster create-or-reuse (with TTL extension) → port exposure → kubeconfig fetch → helm upgrade --install --wait → success/failure status updates
• Implements teardown mode (triggered on PR close) that removes the cluster and posts a final ⬛ comment
• Uses marocchino/sticky-pull-request-comment@v2 with header cmx-helm-preview to avoid conflicts with other actions
• Uses gh api for GitHub deployment status tracking with environment URLs
• Composes existing create-cluster, expose-port, and remove-cluster actions from this repo; inlines helm CLI commands directly (since helm-install is a node24 action and needs kubeconfig as a file, not inline)

Inputs

Input	Default	Notes
replicated-api-token	—	required
chart	—	OCI URL, required
chart-version	—	required
release-name	—	required
namespace	default
values	—	inline YAML, mutually exclusive with values-file
values-file	—	path to file
cluster-name	—	required, e.g. pr-123
cluster-distribution	k3s
cluster-version	1.32
cluster-ttl	24h
port	443
github-token	—	required, needs deployments:write + pull-requests:write
pr-number	—	required
environment-prefix	pr
teardown	false	set true on PR close
cluster-id	—	required when teardown=true

Outputs

• cluster-id — the CMX cluster ID
• cluster-hostname — the ingress hostname from expose-port

Test plan

• Review deploy-helm-preview/action.yml for correctness of step conditionals (teardown != 'true' vs teardown == 'true')
• Verify create-cluster / expose-port / remove-cluster action refs use the correct tag (@v1)
• Validate the cluster-id output expression covers both create and reuse paths
• Confirm marocchino/sticky-pull-request-comment@v2 input naming matches the action's actual interface
• Wire up in a test repo workflow and run against a real PR to exercise the full deploy + teardown flow
• Decide strategy for persisting cluster-id between deploy and teardown jobs (artifact, repo variable, or environment)

🤖 Generated with Claude Code