llama : support Llama 3 HF conversion (#6745)

* Support Llama 3 conversion

The tokenizer is BPE.

* style

* Accept suggestion

Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>

* llama : add llama_token_is_eog()

ggml-ci

* llama : auto-detect more EOT tokens when missing in KV data

* convert : replacing EOS token is a hack

* llama : fix codegemma EOT token + add TODOs

* llama : fix model type string for 8B model

---------

Co-authored-by: Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>