[MinMax] Embedding nodes as input nodes for inference graph

[daniil-lyakhov]

Reopen of the #2862

Changes

• Embedding nodes are used as input nodes for the inference graph (with that embedding nodes are being included to the inference_nncf_graph)
• inference_nncf_graph is used to identify weighted nodes
• PT/FX MinMax get_weight_nodes method is updated to work with the inference graph
• Constant folding is removed from the OpenVINOQuantizer and FX nncf.quantize implementation

Reason for changes

• To prevent quantization of the constant branches without constant folding (Openvino backend for Executorch to enable inference on Intel CPUs, GPUs, NPUs pytorch/executorch#8573 (comment))

Related tickets

163025

Tests

• tests/cross_fw/test_templates/test_quantizer_config.py is updated with shape_of /constant embedding model and conv model with constant branches
• TorchFX reference graphs for VIT and Swin were updated: constant branches are present in the quantized graph but they don't have quantizers inside
• conformance test post_training_quantization/625/ - passed