arg: add --cache-list argument to list cached models

[ngxson]

Important

This PR changes the cached manifest format. You won't see your existing cached models until you update the manifest at least once, by running -hf <user>/<model>:<tag>
(Only the cached manifest will be updated, the GGUF file will NOT be re-downloaded)

Command:

llama-cli -cl

Output:

model cache directory: /Users/REDACTED/Library/Caches/llama.cpp/
number of models in cache: 7
   1. ggml-org/SmolVLM2-500M-Video-Instruct-GGUF:Q8_0
   2. cjpais/llava-1.6-mistral-7b-gguf:Q3_K_M
   3. THUDM/glm-edge-v-5b-gguf:Q4_K_M
   4. ggml-org/gemma-3-4b-it-GGUF:Q4_K_M
   5. second-state/Llava-v1.5-7B-GGUF:Q2_K
   6. ggml-org/SmolVLM2-2.2B-Instruct-GGUF:Q4_K_M
   7. ggml-org/SmolVLM-500M-Instruct-GGUF:Q8_0

[ngxson]

Hmm, just realized that my naming format missed one thing. The naming of manifest file doesn't allow separating username and repo name. For example user_name_repo_name, we don't know if the username is just user or user_name

I think it's better to store the original repo name in the manifest file itself. This will allow user to refresh the manifest in the future. Will do that in a follow-up PR

[ngxson]

I added the new manifest naming format, which should preserve the <user>/<model> structure. This is particularly useful when we want to update or load the model by directly copy-paste the hf_repo name

[ngxson]

@ggerganov Could you have a look at this PR? Thanks!