Fix fp32->fp16 cast for scalars (fixes llama2 fp16 for MPS) (#1752)

Summary:
Summary of changes:
- Fix FP32 -> FP16 scalar conversion
- Disable MPS partitioner by default in mps_example script

```bash
# AOT:
python3 -m examples.apple.mps.scripts.mps_example --model_name="llama2" --bundled

# Runtime:
./cmake-out/examples/apple/mps/mps_executor_runner --model_path llama2_mps_bundled_fp16.pte --bundled_program
```
cc cccclai

Pull Request resolved: https://github.com/pytorch/executorch/pull/1752

Reviewed By: cccclai

Differential Revision: D53236633

Pulled By: shoumikhin

fbshipit-source-id: ab04587bfd1493b81fa03a3145c0df6911749fec