main: crashing upon loading model since commit 83b72cb0 - Windows MSVC + CUDA

[dranger003]

Commit 83b72cb introduces a bug where main.exe crashes immediately after loading the model running on Windows.

The crash occurs in void gguf_free(struct gguf_context * ctx).

cmake -S . -B build -DLLAMA_CUDA=ON && cmake --build build --config Release

main -ngl 33 -c 0 -f prompt.txt -m ggml-meta-llama-3-8b-instruct-f16.gguf

>	llama.dll!gguf_free(gguf_context * ctx) Line 20991	C
 	llama.dll!llama_model_loader::~llama_model_loader() Line 3232	C++
 	llama.dll!llama_model_load(const std::string & fname, llama_model & model, llama_model_params & params) Line 6042	C++
 	llama.dll!llama_load_model_from_file(const char * path_model, llama_model_params params) Line 15225	C++
 	main.exe!llama_init_from_gpt_params(gpt_params & params) Line 2224	C++
 	main.exe!main(int argc, char * * argv) Line 199	C++

[dranger003]

@slaren That was quick :) tested successfully using the PR, thanks!