Wasm: expose WasmAccelerator.TotalKernelsCompiled diagnostic

A proxy for the worker-side _modulesById accumulation (each distinct kernel stays cached as a
compiled WebAssembly.Module on every persistent worker for the tab's life, never evicted). Pairs
with SharedWasmMemoryPages to attribute late-lane process-memory pressure — the lead hypothesis for
the ML heavy-test full-sweep timeouts (held ~3 GiB shared memories + unbounded module cache).
Diagnostic only; no behavior change.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: LostBeard

SpawnDev.ILGPU/Wasm/WasmAccelerator.cs

@@ -303,6 +303,16 @@ public static void DisposeSharedWasmMemory()
         /// </summary>
         private static int _nextKernelId = 0;
 
+        /// <summary>
+        /// Total distinct kernels compiled this session (the monotonic kernelId high-water). DIAGNOSTIC:
+        /// a proxy for the worker-side <c>_modulesById</c> accumulation — each distinct kernel stays
+        /// cached as a compiled <c>WebAssembly.Module</c> on every persistent worker for the tab's life
+        /// (never evicted), so this counts the modules piling up per worker across a long test lane.
+        /// Pair with <see cref="SharedWasmMemoryPages"/> (held shared-memory high-water) to attribute
+        /// late-lane process-memory pressure (the heavy-test full-sweep timeout investigation, 2026-06-14).
+        /// </summary>
+        public static int TotalKernelsCompiled => _nextKernelId;
+
         /// <summary>
         /// Worker-init tracking for one distinct kernel (keyed by its <c>wasmBytes</c> reference in
         /// <see cref="_initializedWorkersByKernel"/>). Carries a stable, unique <see cref="KernelId"/>