WCR: New privacy settings UI is unclear, has potential localization issues

Nobody has claimed this yet.

### Describe the bug

The new "Text and image generation" settings UI in Privacy & Security has usability and localization concerns:

1. Unclear toggle hierarchy: The relationship between the global "Text and image generation" toggle and the app-level "Let apps use Text and image generation" toggle is confusing. It's unclear what the app-level toggle's effective state is when the global toggle is already enabled.

I thought toggling this would allow me to enable AI API access for all apps with and without identity, but that was not the case. I believe the intent of this top-level option in other areas (e.g., Pictures library access) was to give users a master switch to cover apps _without identity_.

2. Improper resource string reuse: The text "Text and image generation" appears to be reused as a substring across multiple UI elements, leading to inconsistent capitalization. In addition to the grammar being wrong in en-US, this pattern will likely cause localization issues in languages with different grammatical constructs where the same phrase may need different forms depending on context.

### Steps to reproduce the bug

1. Install Windows Insider build 26200.5670 or higher (dev channel)
2. Open Settings and go to Privacy & security > Text and image generation
3. Observe UI (screenshot below)

### Expected behavior

_No response_

### Screenshots

### NuGet package version

Windows App SDK 1.8 Experimental 4: 1.8.250702007-experimental4

### Packaging type

Packaged (MSIX)

### Windows version

Insider Build (xxxxx)

### IDE

Visual Studio 2022-preview

### Additional context

Feedback Hub item: https://aka.ms/AAwzf3r

Start by reproducing the UI at Settings > Privacy & security > Text and image generation on the specified Insider build, then inspect the global and app-level toggle behavior and the reused resource text. Done means the toggle hierarchy is unambiguous and the labels use context-appropriate strings that can be localized correctly.

Written by the indexing model from the issue text.