Support loading GGUF model from memory buffer (not just file)

[calebnwokocha]

Problem: Currently, llama_model_load_from_file(...) requires a file path. Llama.cpp uses mmap or reads from disk.

Proposal: Add a function like llama_model_load_from_buffer(const void* buf, size_t size, llama_context_params params) to allow loading GGUF from an in-memory buffer.

Use Cases:

• Embeddable executables that embed the GGUF model as data (not external file).
• Dynamically loaded model blobs (e.g., downloaded into memory, then loaded).