[MHLO] refactor pass configurations

[tanyokwok]

Related to #1227

1. Reduce MHLO #ifdefs
2. Dismiss compilation warnings

[Vremold]

Overall LGTM. Should we also configure enableI32Index and enableStaticShape in torch-backend-to-mhlo-backend-pipeline, which contains convert-torch-to-mhlo pass?

[silvasean]

Thanks!

[tanyokwok]

Overall LGTM. Could we add this feature to torch-backend-to-mhlo-backend-pipeline registered in lib/Dialect/TorchConversion/Transforms/Passes.cpp, too?

I have no idea if it's the right place to register in lib/Dialect/TorchConversion/Transforms/Passes.cpp. What do you think @silvasean @ZihengJiang

[silvasean]

Overall LGTM. Could we add this feature to torch-backend-to-mhlo-backend-pipeline registered in lib/Dialect/TorchConversion/Transforms/Passes.cpp, too?

I have no idea if it's the right place to register in lib/Dialect/TorchConversion/Transforms/Passes.cpp. What do you think @silvasean @ZihengJiang

We already have this registered there:

#ifdef TORCH_MLIR_ENABLE_MHLO
  mlir::PassPipelineRegistration<Torch::TorchLoweringPipelineOptions>(
      "torch-backend-to-mhlo-backend-pipeline",
      "Pipeline lowering torch backend contract to MHLO backend "
      "contract.",
      TorchConversion::createTorchBackendToMhloBackendPipeline);
#endif

You can replace Torch::TorchLoweringPipelineOptions with a new struct with options parallel to def ConvertTorchToMhlo : Pass<"convert-torch-to-mhlo", "func::FuncOp"> { (this is redundant but required in MLIR now).

[ZihengJiang]

Look good to me :)

[tanyokwok]

Okk, @Vremold @silvasean I will split it into another PR.