ENH: Support concurrent embedding, update LangChain QA demo with multithreaded embedding creation

[jiayini1119]

[aresnow1]

Embedding is a CPU-intensive call, and even for a stateless actor, it is not executed simultaneously because the current loop lock is not released until the first call. Therefore, the embedding operation needs to be called with 'to_thread' in model actor. However, I have tried it, and even embedding is not thread-safe for llamacpp, and the process results in a core dump if called concurrently.

[jiayini1119]

We can first try supporting concurrent embedding creation for PyTorch models.

Embedding is a CPU-intensive call, and even for a stateless actor, it is not executed simultaneously because the current loop lock is not released until the first call. Therefore, the embedding operation needs to be called with 'to_thread' in model actor. However, I have tried it, and even embedding is not thread-safe for llamacpp, and the process results in a core dump if called concurrently.