Split mode graph for dense Gemma4 assistant (#2022)

* WIP: Split mode graph for Gemma4 assistant

Something is not right - acceptance drops to nearly zero.

* Per model CUDA contexts

Still not working!?

* This works

The issue was that I was not correctly calculating the number
of KV heads for the split KV cache.

* Compiler warnings

* It is better to use llama_context pointers as keys

examples/cvector-generator/pca.hpp

@@ -66,7 +66,7 @@ struct pca_model {
     pca_model(struct ggml_tensor * t_input) {
 #ifdef GGML_USE_CUDA
         fprintf(stderr, "%s: using CUDA backend\n", __func__);
-        backend = ggml_backend_cuda_init(0, nullptr); // init device 0
+        backend = ggml_backend_cuda_init(0, nullptr, nullptr); // init device 0
         if (!backend) {
             fprintf(stderr, "%s: ggml_backend_cuda_init() failed\n", __func__);
         }