ggml-alloc : allocate all leafs as if they were inputs (#731)

ggml-alloc : allocate all leafs as if they were inputs (#731)

* ggml-alloc : allocate all leafs as if they were inputs

* ensure static leafs are allocated

* gpt-2-backend : remove unnecesary ggml_new_tensor

* update other gpt-2 examples to remove ggml_new_tensor calls in the graph