Integrate DCG with command auto-approval and denied-command UX

[navedmerchant]

Parent

Part of #1049. This is the fourth and final independently reviewable slice of the Destructive Command Guard work.

Dependency

Depends on the persisted DCG setting and service slices.

Problem

When command auto-approval and DCG are enabled, safe complex/generated commands should run without the legacy allowlist interrupting them, while DCG-blocked commands must retain an explicit approval boundary and clear denial feedback.

Scope

• Evaluate commands with DCG before asking for command approval.
• Allow DCG-approved commands to bypass the legacy allowlist only when command auto-approval and DCG are both enabled.
• Mark DCG-blocked commands as protected so they always require explicit user approval.
• Preserve existing allowlist/denylist behavior when DCG is disabled.
• Record auto-deny decisions in command messages and render a dedicated denied state in chat.
• Show localized DCG reason/rule details when available and hide allow/deny pattern controls where they do not apply.
• Add focused extension and webview regression tests for policy precedence, command execution, message persistence, and denied rendering.

Acceptance criteria

• DCG-approved commands auto-run only when command auto-approval is enabled.
• DCG-blocked commands cannot cross the explicit approval boundary automatically.
• User approval can still execute a DCG-blocked command; user denial does not execute it.
• DCG-disabled behavior is unchanged.
• Denied commands render a clear localized status and optional reason/rule metadata.
• Focused command-policy, tool, message, and chat tests pass.
• The complete stacked branch type-checks/builds and the combined feature runs.

Stack

Base: the persisted-setting PR. This final PR completes #1049.