Add chatml fallback for cpp llama_chat_apply_template

[ngxson]

Fix problem with DeepSeek V2 chat model: #8068 (comment) (cc @fairydreaming )

DeepSeek is no longer crash now, but it's using chatml (not ideal). We can add template for deepseek later.

Demo:

make llama-cli && ./llama-cli -m ../DeepSeek-Coder-V2-Lite-Instruct-Q2_K.gguf -cnv -p "You are Bob" -c 256

> hi
Hello! How can I assist you today?

> who are u
I am an AI assistant developed by DeepSeek Company. I can help you answer questions and provide information. How can I assist you today?

>

---

• I have read the contributing guidelines
• Self-reported review complexity:
  • Low

[fairydreaming]

DeepSeek-V2-Lite no longer crashes with these fixes, but I think there's one unnecessary llama_chat_apply_template() call that can be removed.

[fairydreaming]

Is this really necessary? I mean you already called llama_chat_apply_template(nullptr, "chatml", ...) in else above, so why do you call it again?

[ngxson]

Yeah that's right, I forgot to delete the line inside else condition

[ngxson]

I deleted this line of code (keep the line in else branch). I've tested it once more time and confirmed that it's still working.

This will be merged once CI passed.