Enable and Disable Copilot features at the semantic model (dataset) level

[Nightwolf_112]

Problem:

• Copilot controls are still too broad (tenant, capacity and workspace, AD group).
• Even with a separate smaller Copilot capacity to redirect compute away from main capacity, individual heavy models can still drive compute and cost inside that capacity leading to rejection states. Large or sensitive models feel this the most, especially when chat and prompts hit DirectQuery or composite designs.
• If we turn Copilot off at the org, workspace, or AD-group level, we also lose useful features like AI summaries in subscriptions.
• We need guardrails that protect fabric capacity during peak periods without shutting Copilot off everywhere.

Suggestion:
Add a per semantic model setting that lets model owners enable or disable specific Copilot capabilities for that model, independent of tenant or workspace defaults. This allows precise control where compute and cost or performance constraints are highestwhile keeping benefits like AI summaries available even if chat is turned off.

Examples:

Semantic Model A: turn Copilot off completely.

Semantic Model B: keep AI summaries for subscriptions but disable Copilot chat.

We should expose Copilot only on models that have been prepared and performance validated, and keep it off by default for everything else.