Misc. bug: strange reducing memsize type to 32bit without dev comment

[GermanAizek]

Contributors, What's the point of truncating to a strict 32-bit size_t value and comparing to int64_t? Is this a legacy cast-code when it was rewritten to 64-bit types, or is that the way it was intended?

https://github.com/ggerganov/llama.cpp/blob/a1649cc13f89946322358f92ea268ae1b7b5096c/src/llama-sampling.cpp#L394-L396