[core] Support 2D Bias for Gemm operator
According to ONNX's [documentation](https://onnx.ai/onnx/operators/onnx__Gemm.html#summary), the Bias should be broadcastable to the shape (BatchSize, OutputFeature) which means it could be given with that shape.
This is already the case for one of the ONNX [tests](https://github.com/onnx/onnx/blob/v1.16.2/onnx/backend/test/data/node/test_gemm_default_matrix_bias/model.onnx):
issue