When quantising models and overriding integer metadata parameters at the same time, their assigned type becomes int even though their original value is unsigned int. In certain cases this behaviour triggers an exception when loading the quantised model.

For example, using the model available here:

1. Quantise & override: llama-quantize --override-kv qwen3moe.expert_used_count=int:16 Qwen3-30B-A3B-BF16.gguf Qwen3-30B-A3B-Q4_K_M.gguf Q4_K_M
2. Load model: llama-simple -m Qwen3-30B-A3B-Q4_K_M.gguf "Hello, world!"

Will lead to an error loading model hyperparameters: key qwen3moe.expert_used_count has wrong type i32 but expected type u32 exception

This PR changes the if (params->kv_overrides) logic in llama-quant.cpp to use uint32 if there are any int overrides, so that llama-quantize --override-kv qwen3moe.expert_used_count=int:16 Qwen3-30B-A3B-BF16.gguf Qwen3-30B-A3B-Q4_K_M.gguf Q4_K_M generates a functioning model

More context here