[Task] Windows Copilot Runtime integration (on-device summarization)

[shellen]

Part of: [Epic] Add Windows support (#66)

Summary

Add a windows-ai provider to our LLM provider list for on-device summarization on Copilot+ PCs, using the Windows AI Foundry (formerly Windows Copilot Runtime).

Platform Details

• Windows App SDK 1.7+ (shipped mid-2025) includes Phi Silica (~3.3B params, NPU-optimized)
• TextSummarizer API: Microsoft.Windows.AI.Generative namespace
• Availability: Copilot+ PCs only (Snapdragon X, Intel Core Ultra, AMD Ryzen AI) — requires NPU hardware
• API pattern: Similar to Apple Intelligence — check availability, create session, send text, get summary

Implementation Approach

Add a small Rust or C# helper invoked from the Tauri backend (similar to how we shell out for Apple Intelligence via osascript). Gate on TextSummarizer.IsAvailable().

if (TextSummarizer.IsAvailable()) {
    var summarizer = await TextSummarizer.CreateAsync();
    var result = await summarizer.SummarizeAsync(articleText);
}

Acceptance Criteria

• windows-ai provider added to LLM provider list
• Provider gated on hardware availability check
• Summarization works on Copilot+ PC hardware
• Graceful fallback when NPU hardware not available

Estimate: ~2 days
Priority: Nice-to-have