[Guidance] Document component requirements and cost-optimization options

### Problem

Organizations deploying GPT-RAG need clearer guidance on which infrastructure components are required for their deployment scenario, which are optional, and which are used only during setup or ongoing administration.

Without that distinction, teams may retain unnecessary resources or remove components that are still needed for networking, security, or operations.

### Proposed improvement

Add a component reference covering resources such as Application Gateway, API Management, Azure Bastion, virtual machines, and NAT Gateway. For each component, document:

- Its purpose, dependencies, and the scenarios in which it is required.
- Whether it supports provisioning, application runtime, or ongoing operations.
- Supported configuration options for omitting or disabling it.
- Conditions for safe decommissioning, including security, connectivity, and infrastructure-as-code implications.
- Relevant cost drivers and component-level optimization options.

Keep this focused on component selection and lifecycle. Workload-based sizing profiles and full deployment cost examples belong in the separate sizing guidance issue.

### Expected outcome

Teams can select an appropriate configuration and understand which resources must remain after deployment.

### Related

Azure/GPT-RAG#693 covers workload-based sizing profiles and deployment cost estimates.

No files or tests are named. Start by locating the existing GPT-RAG deployment and infrastructure-as-code definitions for Application Gateway, API Management, Azure Bastion, virtual machines, and NAT Gateway; compare them with the separate sizing guidance in issue 693. Done means a component reference explains purpose, lifecycle, omission or disabling options, safe decommissioning conditions, dependencies, and cost drivers without duplicating workload sizing guidance.

Written by the indexing model from the issue text.