tests: benchdnn: graph: remove legacy sdpa bwd cases

Author: ElaineBao

tests/benchdnn/inputs/graph/complex_fusion/harness_mha_all

@@ -33,8 +33,6 @@
 --reset --dt=0:f16+1:f16+7:f16+9:f16+10:f16 --case=complex_fusion/mha/sdpa-plain-wo-mask-f32.json
 --reset --dt=0:f16+1:f16+7:f16+8:f16+9:f16 --case=complex_fusion/mha/sdpa-plain-wo-scale-f32.json
 --reset --in-shapes=0:2x1x1024x128*131072x128x128x1+1:2x1x1024x128*131072x128x128x1+2:1*1+5:2x1x1024x128*131072x128x128x1 --case=complex_fusion/mha/sdpa-training-fwd-no-mask-f16-f32.json
---reset --dt=16:f16+17:f16+32:f16+33:f16+34:f16+36:f16+44:f16+45:f16+47:f16 --case=complex_fusion/mha/sdpa-plain-training-backward-w-dmask-bf16-f32.json
---reset --dt=16:f16+17:f16+20:f16+32:f16+33:f16+34:f16+36:f16+38:f16+46:f16+47:f16+49:f16+51:f16 --case=complex_fusion/mha/gqa-plain-training-backward-w-dmask-bf16-f32.json
 
 # bf16 inputs + f32 intermediates + bf16 outputs
 --reset --op-kind=1:Multiply,1:Divide --dt=1:bf16+2:bf16+3:bf16+4:bf16+5:bf16+6:bf16+104:bf16 --case=complex_fusion/mha/sdpa-plain-simplified-f16-f32.json
@@ -53,8 +51,6 @@
 --reset --dt=1:bf16+2:bf16+3:bf16+4:bf16+6:bf16+104:bf16 --case=complex_fusion/mha/sdpa-plain-implicit-causal-mask-fp32-bs1.json
 --reset --dt=0:bf16+1:bf16+7:bf16+9:bf16+10:bf16 --case=complex_fusion/mha/sdpa-plain-wo-mask-f32.json
 --reset --dt=0:bf16+1:bf16+7:bf16+8:bf16+9:bf16 --case=complex_fusion/mha/sdpa-plain-wo-scale-f32.json
---reset --case=complex_fusion/mha/sdpa-plain-training-backward-w-dmask-bf16-f32.json
---reset --case=complex_fusion/mha/gqa-plain-training-backward-w-dmask-bf16-f32.json
 
 # int8 graphs
 --reset --case=complex_fusion/mha/MHA-GPT-inf-int8-bs1.json

tests/benchdnn/inputs/graph/complex_fusion/harness_mha_ci

@@ -33,7 +33,6 @@
 --reset --dt=0:f16+1:f16+7:f16+9:f16+10:f16 --case=complex_fusion/mha/sdpa-plain-wo-mask-f32.json
 --reset --dt=0:f16+1:f16+7:f16+8:f16+9:f16 --case=complex_fusion/mha/sdpa-plain-wo-scale-f32.json
 --reset --in-shapes=0:2x1x1024x128*131072x128x128x1+1:2x1x1024x128*131072x128x128x1+2:1*1+5:2x1x1024x128*131072x128x128x1 --case=complex_fusion/mha/sdpa-training-fwd-no-mask-f16-f32.json
---reset --dt=16:f16+17:f16+20:f16+32:f16+33:f16+34:f16+36:f16+38:f16+46:f16+47:f16+49:f16+51:f16 --case=complex_fusion/mha/gqa-plain-training-backward-w-dmask-bf16-f32.json
 
 # bf16 inputs + f32 intermediates + bf16 outputs
 --reset --op-kind=1:Multiply,1:Divide --dt=1:bf16+2:bf16+3:bf16+4:bf16+5:bf16+6:bf16+104:bf16 --case=complex_fusion/mha/sdpa-plain-simplified-f16-f32.json
@@ -45,15 +44,13 @@
 --reset --dt=0:bf16+1:bf16+3:bf16+7:bf16+2:bf16+8:bf16 --case=complex_fusion/mha/MHA-stable_diffusion-inf-fp32-bs1.json
 --reset --case=complex_fusion/mha/sdpa-plain-training-forward-bf16-f32.json
 --reset --case=complex_fusion/mha/sdpa-plain-training-backward-bf16-f32.json
---reset --case=complex_fusion/mha/sdpa-plain-training-backward-w-dmask-bf16-f32.json
 --reset --case=complex_fusion/mha/gqa-plain-training-forward-bf16-f32.json
 --reset --case=complex_fusion/mha/gqa-plain-training-backward-bf16-f32.json
 --reset --case=complex_fusion/mha/codegemma-bf16-f32.json
 --reset --case=complex_fusion/mha/gemma2-bf16-f32.json
 --reset --dt=1:bf16+2:bf16+3:bf16+4:bf16+6:bf16+104:bf16 --case=complex_fusion/mha/sdpa-plain-implicit-causal-mask-fp32-bs1.json
 --reset --dt=0:bf16+1:bf16+7:bf16+9:bf16+10:bf16 --case=complex_fusion/mha/sdpa-plain-wo-mask-f32.json
 --reset --dt=0:bf16+1:bf16+7:bf16+8:bf16+9:bf16  --case=complex_fusion/mha/sdpa-plain-wo-scale-f32.json
---reset --case=complex_fusion/mha/gqa-plain-training-backward-w-dmask-bf16-f32.json
 
 # int8 graphs
 --reset --case=complex_fusion/mha/MHA-GPT-inf-int8-bs1.json