ggml : ggml_graph_compute should not require ggml_context

[ggerganov]

We now have the following signature:

    GGML_API void ggml_graph_compute(struct ggml_context * ctx, struct ggml_cgraph * cgraph);

The provided context is used during the computation to potentially allocate a work buffer needed by some of the ggml operators. Not only the buffer is not always needed, but having to pass an entire context is a poor design choice.

We need to avoid this by allowing the user to pass a work buffer (pointer and size) and an API that allows the user to query what is the work size needed to evaluate a specific compute graph. This way, the user will first make the query and then will be responsible to provide the work buffer externally if necessary