ggml-openvino : simplify op translators and enable IMROPE/NEOX RoPE fusion

[mostafafaheem]

Overview

1. Simplifying and cleaning up op translators of MUL_MAT, DIAG, GEGLU_QUICK, SWIGLU, SWIGLU_CLAMP, SWIGLU_OAI, ROPE, SILU and SOFTPLUS, using native OV ops when possible.
2. Rewriting NEOX and IMROPE translators to align with ov RoPE fusion on CPU, GPU, NPU. They now emit ov::op::internal::Rope

Additional information

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes, inefficiency analysis and some of the implementation

[wine99]

Changes look good to me. But I have not built and verified the models

[mostafafaheem]

Changes look good to me. But I have not built and verified the models

I ran llama-bench on the 34 models in using a batch script on CPU, GPU and NPU, stateful and stateless, and there is no difference in model support