A Selective Survey of Efficient Speculative Decoding Techniques for LLM Inference - eviltoast