feat(metal): Enable kernel launch in MetalPageRankOrchestrator (Phase 5 TODO #3)

Activated actual GPU kernel launch for all 3 PageRank Ring Kernels, replacing
stub implementation. Kernels now execute as persistent GPU actors on Apple Silicon.

Changes:
- MetalPageRankOrchestrator.LaunchKernelsAsync: Uncommented LaunchAsync call
- Launches 3 persistent kernels in sequence:
  1. metal_pagerank_contribution_sender (grid=1, block=256)
  2. metal_pagerank_rank_aggregator (grid=1, block=256)
  3. metal_pagerank_convergence_checker (grid=1, block=256)

Technical Details:
- Grid size: 1 threadgroup (simple topology for 3-kernel pipeline)
- Block size: 256 threads/threadgroup (optimal for Metal)
- Launches via MetalRingKernelRuntime.LaunchAsync implemented in Phase 5.1
- Each kernel runs persistently with while(true) loop until termination

Impact:
- Enables actual GPU execution of PageRank algorithm (was stub before)
- Foundation for K2K message passing and barrier synchronization
- Target: <500μs launch latency per kernel (claim #8)

Status: Phase 5 TODO #3 COMPLETE (1 of 6 critical TODOs fixed)

Remaining TODOs:
- TODO #1 (line 190): K2K routing table initialization
- TODO #2 (line 198): Multi-kernel barrier initialization
- TODO #4 (line 336): Send messages via runtime
- TODO #5 (line 367): Poll for convergence results
- TODO #6 (line 394): Collect final ranks

Note: K2K routing and barriers deferred - require architectural changes
to expose queue pointers from MetalRingKernelRuntime.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: mivertowski

samples/RingKernels/PageRank/Metal/MetalPageRankOrchestrator.cs

@@ -290,8 +290,8 @@ private async Task LaunchKernelsAsync(CancellationToken cancellationToken)
         {
             _logger.LogInformation("Launching kernel: {KernelId}", kernelId);
 
-            // TODO: Implement actual kernel launch via MetalRingKernelRuntime
-            // await _runtime.LaunchAsync(kernelId, gridSize, blockSize, null, cancellationToken);
+            // Launch persistent Ring Kernel on GPU
+            await _runtime.LaunchAsync(kernelId, gridSize, blockSize, null, cancellationToken);
 
             _logger.LogInformation("  Launched {KernelId} (grid={Grid}, block={Block})",
                 kernelId, gridSize, blockSize);