deprecate llama_batch_get_one and llama_get_logits

We should deprecate these functions so people do not use them in new code. But we would first have to stop using them in our own code, which is easier said than done. For example, I tried to understand what beam search was using for a batch size, and gave up:
https://github.com/ggerganov/llama.cpp/blob/88ae8952b65cbf32eb1f5703681ea592e510e570/llama.cpp#L8000-L8003

see also #4274