[ONNX] Expand more ONNX operators as functions when importing

[andfau-amd]

https://github.com/llvm/torch-mlir/pull/3409 / https://github.com/llvm/torch-mlir/commit/51902ec2dc6df99a87e0fee092e59b492ab04837 introduced the ability for the ONNX Python importer to expand ONNX operators that are defined as functions to MLIR functions. This is only done for allowlisted operators, and at the time of writing, there's only one: MeanVarianceNormalization.

This issue is for tracking/triaging potential additions to the allowlist:

• Operators without existing TorchOnnxToTorch conversions that could make use of function expansion.
• Operators with existing TorchOnnxToTorch conversions that could be removed in favour of function expansion.

[andfau-amd]

https://github.com/llvm/torch-mlir/issues/3527 is relevant