Add ffn_gate_up_exps to --cpu-moe and --n-cpu-moe overrides

[ikawrakow]

Haha, lots of little details around the merged ffn_gate_up_exps tensors imposed on the world by llama.cpp maintainers.

This PR adds tensor overrides for these tensors when using --cpu-moe or --n-cpu-moe.

Thanks to @ubergarm for reminding me with this comment.

[ubergarm]

oh lol i was just working on this PR after seeing it break on mainline yesterday: ggml-org/llama.cpp#20416

so we're ffn_gate_up_exps everywhere now, so it is

thanks and happy weekend!