fix: decouple dashboard deploys from retired indexer state

[ludamad]

Why

The testnet dashboard deploy failed because it still read cf_domain_name from the retired testnet-green ATP indexer state. That state was intentionally destroyed during the single-instance migration, so Terraform now sees an output-less state object and aborts before building the frontend.

Fix

• use the stable branded API endpoints directly for migrated prod/testnet dashboards
• skip the retired indexer remote state for those environments
• make testnet the valid default for manual and testnet-tag deploys
• update the single-instance cutover docs to match

Verification

• merged-main Build workflow passed
• terraform validate passed
• real testnet plan passed with indexer_deployment_suffix=-green
• plan changes zero AWS resources; only atp_indexer_url changes from the retired CloudFront hostname to https://api.testnet.stake.aztec.network
• live testnet API /ready and /api/providers return 200