Integrate sdk (#2314)

Summary:
The PR adds support for logging profiling information and intermediate outputs. The required files are only included if CoreML delegate is built with`EXECUTORCH_BUILD_SDK` option.

- Profiling:  `ETCoreMLModelProfiler` adds profiling support, it uses the `MLComputePlan` API that's currently in beta but unlikely to change. The caveat is that it only works for macOS >= 14.4, iOS >= 17.4.
- Debugging: `ETCoreMLModelDebugger` adds support for getting intermediate tensors. The model spec is updated to include additional intermediate outputs, compiled, and then executed. It has a dependency on `protobuf` as it modifies the specification of the model. The logging part doesn't work yet, it would require ExecuTorch API (`log_tensor_delegate`).
- CLI: Added cli file for parsing CoreML delegate metadata.
- AOT: AOT integration is remaining but most likely it would be done in `coremltools`. `coremltools` will add a `json` file to `mlpackage` that would contain the operation path to debug handle mapping.

![Screenshot 2024-03-07 at 11 04 00 PM](https://github.com/pytorch/executorch/assets/709448/62bbc623-cd53-42cd-bec7-f7afca227f79)

Pull Request resolved: https://github.com/pytorch/executorch/pull/2314

Reviewed By: cccclai

Differential Revision: D54754857

Pulled By: shoumikhin

fbshipit-source-id: 493da4bfb8f9a4d60d68a219cfed63178d731add