fix: EC3 v3 image registry routing

[aa-replicated]

Summary

• Remove imageProxy.host conditional from deployment.yaml — template now always uses image.registry/image.repository:tag directly
• Update values.yaml image defaults to full proxy URL format (proxy.replicated.com + proxy/gameshelf/ghcr.io/aa-replicated/gameshelf) so CMX installs work without extra --set flags
• Fix helmchart.yaml for EC3 v3: clear imageProxy.host, pass full image refs to ReplicatedImageRegistry (so it returns the correct full proxy prefix, not just the proxy host), and remove all ReplicatedImageRepository overrides (undocumented function)

Root cause of original bug: ReplicatedImageRegistry "ghcr.io" (just a registry host) returned proxy.replicated.com. The deployment template then embedded this in proxy.replicated.com/proxy/gameshelf/<value>/..., doubling the proxy host. The fix passes full image refs so the function has context to return the full prefix.

Test plan

• Promote release in vendor portal and deploy to online EC3 cluster — all 4 pods (gameshelf, postgresql, redis, replicated SDK) reach Running
• Verify no doubled proxy URL in image refs (kubectl describe pod should show proxy.replicated.com/proxy/gameshelf/..., not proxy.replicated.com/proxy/gameshelf/proxy.replicated.com/...)
• If airgap test is available: verify pods pull from local embedded registry
• CMX (Helm CLI) install: helm template output shows correct proxy URLs with default values

🤖 Generated with Claude Code