Created by: jczaja

Changes discussed here are adjusting mkl-dnn ops (no INT8 and elementwise_mul yet) to Tensor modifications introduced in #15429

Scope of Changes:

• set_layout() + set_format() was replaced with set_mkldnn_prim_desc()
• In softmax set_layout() + set_format() were missing hence set_mkldnn_prim_desc() was added

Notes:

• Elementwise_mul and INt8 ops (conv int8, pooling, quantize, dequantize) will be modified in next PR
• Performance should not change with those changes. If you find some performance/functional problems after introducing this change please let us know.