Qwen3 MoE should also work with tie_word_embeddings

[estibi]

For custom trained Qwen3 MoE model with tie_word_embeddings=True I get this error:

load_tensors: loading model tensors, this can take a while... (mmap = true)
llama_model_load: error loading model: missing tensor 'output.weight'
llama_model_load_from_file_impl: failed to load model

This fix makes tie_word_embeddings optional for Qwen3 MoE.