allow head_dim for llama like gemma or mistral

[bzantium]

What does this PR do?

• allow head_dim not to depend on hidden_size for llama like gemma or mistral
• it enables more flexible model hyper parameters without disturbing original models
• refer to: [mistral] Support passing head_dim through config (and do not require head_dim * num_heads == hidden_size) #32050

Fixes #32846

Who can review?

@ArthurZucker