[CPU]support glm4 rope

[zhangYiIntel]

Details:

• Support Rope kernel of GLM4
• the input data order has changed from ([seq_length, batch, 4608]) in ChatGLM3 to ([batch, seq_length, 4608]) in ChatGLM4. Within RoPE process, the data order changes from ([seq_length, batch, head_count, head_size]) to ([batch, head_count, seq_length, head_size]) by permute operation added in ChatGLM4.
• the RoPE cache data order has changed from ([seq_length, batch, head_count, 2]) in ChatGLM3 to ([batch, head_count, seq_length, 2]) in ChatGLM4.
• Consequently, the output of RoPE has also changed from ([seq_length, batch, head_count, head_size]) in ChatGLM3 to ([batch, head_count, seq_length, head_size]) in ChatGLM4
• Due to these changes, the RoPE pattern matching needs to create something new, something different from what already existed ChatGLM pattern matching. Additionally, new kernels need to be added to accommodate these changes

Tickets:

• ticket-id

[luo-cheng2021]

LGTM.