Depend on torch-mlir via CMake

Moving discussion from Discord thread by @garymm

Hi I'm trying to write some C++ code that depends on torch-mlir. Does anyone have an example CMake project that depends on the torch-mlir code? If not, any tips as to the general thing I should be trying to do? I have basically no experience with LLVM's build system. So far I've tried this:
transforms/CMakeLists.txt:

add_mlir_library(transforms
  passes.cpp

  DEPENDS
  TorchMLIRConversionPassIncGen

  LINK_COMPONENTS
  Core

  EXCLUDE_FROM_LIBMLIR
)

transforms/passes.cpp:

#include "mlir/Transforms/Passes.h"
#include "mlir/Pass/PassManager.h"
#include "foo/transforms/passes.h"

namespace foo {

void registerConversionPasses() {
  mlir::PassPipelineRegistration<
      mlir::torch::Torch::TorchLoweringPipelineOptions>(
      "torch-backend-to-foo-backend-pipeline",
      "Pipeline lowering torch backend contract to foobackend "
      "contract.",
      createTorchBackendToFooBackendPipeline);
}

void createTorchBackendToFooBackendPipeline(
    OpPassManager &pm,
    const mlir::torch::Torch::TorchLoweringPipelineOptions &options) {
}

} // namespace foo

The build fails compiling passes.cpp because it can't find the mlir/Transforms/Passes.h to include.

Start with transforms/CMakeLists.txt and transforms/passes.cpp, then inspect how torch-mlir and MLIR dependencies are exposed to external CMake projects. Done means a C++ example project can depend on torch-mlir and compile the shown include successfully.