graph: backend: dnnl: backend refactor and sdpa v1 kernel support quantize SDPA

[xiang1guo]

Background

This is a follow-up work based on #2930 and #2931. The PR mainly focuses on supporting quantized SDPA with internal dnnl_sdpa. It helps to reduce graph compilation time and also simplifies the backend optimization pass.

Works

• DNNL backend refactor:
  • attach fusion info to op attr directly
  • rename fusion_info_mgr to reflect it's current usage
• Support compressed SDPA with internal dnnl_sdpa.
• Support legacy GQA pattern with internal dnnl_sdpa.
• Merge sdp_primitive_kernel_t and sdp_primitive_v1_kernel_t

TODO

• Support CPU decompose kernel with internal dnnl_sdpa.

Testing results:

For all 218 mha test cases, we now have 67 ukernel-optimized cases that can run successfully in the sdp_primitive_v1_kernel_t kernel.

• compressed SDPA dot graph:
  before fusion
  
  after fusion
  

• legacy GQA dot graph
  before fusion
  
  after fusion
  

[TaoLv]

Can you please split the fusion info refactor into a separate PR, for better review experience?

[xiang1guo]

Can you please split the fusion info refactor into a separate PR, for better review experience?

ok, sure, will do that.