Link NVIDIA compatibility failures to driver remediation

Problem

The Local AI setup UI explains why NVIDIA compatibility checks failed, but it does not provide an actionable driver installation or repair link.

Current failure messages cover:

• No NVIDIA GPU reported by the driver
• Missing stable GPU, memory, driver, or CUDA evidence
• NVIDIA driver below the required version
• Insufficient CUDA driver API support

https://github.com/openclaw/openclaw-windows-node/blob/31dffdb0dbd9eebc8b98ce35cf13ab60762410a9/src/OpenClaw.SetupEngine.UI/Pages/CapabilitiesPage.xaml.cs#L390-L414

The unavailable-details control opens explanatory content, but there is no NVIDIA remediation destination:

https://github.com/openclaw/openclaw-windows-node/blob/31dffdb0dbd9eebc8b98ce35cf13ab60762410a9/src/OpenClaw.SetupEngine.UI/Pages/CapabilitiesPage.xaml#L141-L149

Impact

A user who fails an NVIDIA prerequisite receives a diagnosis but must independently search for the correct driver. This creates avoidable setup abandonment and increases the risk of downloading an incorrect or untrusted package.

The managed llama-server runtime requires a compatible NVIDIA display driver. A separate CUDA Toolkit installation is not required.

Expected behavior

• Provide a clear driver installation or repair action in the Local AI unavailable details.
• Use a stable NVIDIA-owned redirect so the destination can change without an application update.
• Show the action for missing, unreadable, outdated, and insufficient-CUDA driver states.
• Do not direct users to install the CUDA Toolkit.
• Keep the action distinct from retrying hardware detection.
• Localize the surrounding copy and provide an accessible name and description.

Tests

Add UI and presentation coverage for:

• Link visibility for each applicable NVIDIA failure state
• Link hidden for unrelated failures such as unsupported Windows architecture or unknown model
• Expected allowlisted HTTPS destination
• Accessible name and keyboard activation
• Retry remains available after remediation

Issue found by @shanselman.

Start with the NVIDIA failure handling in src/OpenClaw.SetupEngine.UI/Pages/CapabilitiesPage.xaml.cs around lines 390-414 and the unavailable-details control in CapabilitiesPage.xaml around lines 141-149. Trace how each failure state reaches the UI, then add UI and presentation coverage for NVIDIA visibility, unrelated-state hiding, the allowlisted HTTPS destination, accessibility, keyboard activation, and retry availability.