[V1] Add code dataset to benchmark the performance of spec decode

[LiuXiaoxuanPKU]

Current we don't have a code-related dataset in the benchmark. It will be nice to have a code editing dataset to demonstrate the major use case and benefits of ngram speculative decoding.

[CXIAAAAA]

#15303 @LiuXiaoxuanPKU @WoosukKwon pls take a look

[CXIAAAAA]

@LiuXiaoxuanPKU this is completed and we could mark it as done