cli : auto activate conversation mode if chat template is available

[ngxson]

The goal of this PR is to provide an automatic way to enable -cnv whenever chat template is available (either built-in or via --chat-template)

Please note that, this is not related to the recent discussion on chat UX (#11203), but rather focus on OOTB experience.

With this PR, the OOTB experience will become:

# install
brew install llama.cpp

# use local gguf file
llama-cli -m Qwen2.5-7B-Instruct-IQ2_M.gguf

# use HF hosted model
llama-cli -hf bartowski/Llama-3.2-1B-Instruct-GGUF

User can still force-disable it via -no-cnv

This also fixes #11157 since it's not obvious that some models need system message.

[ggerganov]

I think ggml-ci needs to be updated after this change: https://github.com/ggml-org/ci/tree/results/llama.cpp/84/a44815f704aaed8e8edec7a39e846a975c7ba9/ggml-2-x86-cpu

[strawberrymelonpanda]

For what it's worth, this was a fairly problematic change for me as my CLI scripts suddenly broke, due to Llama-CLI now deciding that my prompts (passed with -p) are now system prompts instead because of auto-conversation.

I just have to pass -no-cnv to resume script usage, which isn't a big deal, but was quite confusing out of the blue. As a suggestion, perhaps a message under == Running in interactive mode. == that "Conversation is now the default mode; pass "--no-conversation" for scripted usage" would be prudent?

[strawberrymelonpanda]

Also maybe worth mentioning the "Example usage" at the bottom of llama-cli --help could use updating.

example usage:

  text generation:     ./build/bin/llama-cli -m your_model.gguf -p "I believe the meaning of life is" -n 128

  chat (conversation): ./build/bin/llama-cli -m your_model.gguf -p "You are a helpful assistant" -cnv

The first example is no longer text gen, but conversation. The readme adds "-no-cnv" to this. The second example would work, but is redundant, and as per the readme change would need a template if it's being manually specified for an unsupported model.