Add curl support

[danielholanda]

Description

This PR:

• Enables LLAMA_CURL=ON for both Windows and Ubuntu builds
• Windows: Install curl via vcpkg, include libcurl.dll in distribution
• Ubuntu: Install libcurl4-openssl-dev build dependency
• Adds curl functionality tests using -mu (model URL) flag to verify that the direct HuggingFace model downloads works

This allows users to download GGUF models directly from URLs (e.g., HuggingFace) using llama-cli -mu without needing to pre-download files.

To test, run

llama-cli.exe -mu "https://huggingface.co/unsloth/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B-Q4_0.gguf" -ngl 99 -p "Hello" -n 5 -st

Closes #6