graph: enable quantized gated mlp dispatch

[TaoLv]

1. The example and test files in benchdnn are updated with f32 intermediate data type. (Need Arch review.)
2. Graph backend gated mlp kernel to support quantized inputs.
3. Dispatch quantized gated mlp patterns to the quantized gated mlp primitive.

Still, the dispatching is disabled by default, to enable it, set _ONEDNN_GRAPH_GATED_MLP_FORCE_PRIMITIVE=0.

[TaoLv]

make test
disable benchdnn_all
enable benchdnn_graph

[TaoLv]

make test
disable benchdnn_all
enable benchdnn_graph