'infra single-project' fails when run after `deploy` for agent_runtime

### Problem

When using **Agent Runtime** (Reasoning Engine), running `agents-cli infra single-project` after an initial `agents-cli deploy` does not work as intended.

`infra single-project` uses terraform, but `deploy` does not and uses an API instead. This means that when `single-project` tries to use terraform to update, it doesn't have visibility into the existing reasoning engine resource already created. So, the `terraform apply` invoked by `single-project` silently provisions a **second** Reasoning Engine instance with the same name.

This leads to:
- Observability environment variables being wired to the new (empty) instance.
- The original (serving) instance remaining without observability configuration.

### Workarounds

See https://github.com/google/agents-cli/blob/main/skills/google-agents-cli-observability/SKILL.md#order-of-operations-for-agent_runtime-deployments

The short version though: either run `single-project` before `deploy` so tracking is properly established, or fix it manually using the above instructions.

Ref: b/504375097

まず `agents-cli deploy` と `agents-cli infra single-project` のフローから始め、続いて `skills/google-agents-cli-observability/SKILL.md` にある操作順序のガイダンスを読んでください。その順序でコマンドを再現し、`single-project` が 2 つ目のインスタンスを作成するのではなく既存の Reasoning Engine を更新し、serving インスタンスに observability 変数が付加されていることを確認してください。

索引モデルが issue の本文から書いたものです。