[Feature] Allow async model loading and cancellation

[SanftMonster]

In production environment, especially desktop apps, it's common to have a button (or any other way) to allow users to abort the model loading. Fortunately, llama.cpp has already added support for it ggml-org/llama.cpp#4462. I think we should introduce this feature in LLamaSharp.

Similarly, async model loading is also important for applications based on LLamaSharp, which avoids blocking the main thread for a long time when loading a large model. I've found a similar work of it in the node.js binding of llama.cpp withcatai/node-llama-cpp#178. We could also implement it by polling the progress callback.