Parallelize SPIR-V compilation (#4200)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/4200

Due to a high number of shaders to compile, shader compilation has become quite slow (taking around 1 minute from empirical observations). This change updates the SPIR-V compilation script to parallelize SPIR-V compilation over the available number of CPU cores.

Testing the impact of parallelization using the following command:

```
buck build //xplat/executorch/backends/vulkan:gen_vulkan_graph_runtime_shaderlib_cpp --out ~/scratch/shaders
```

We can observe the following improvement:

| | Before | After|
|--- | --- | --- |
| M1 Mac | 42.4s | 7.8s |

Reviewed By: liuk22, jorgep31415

Differential Revision: D59597186

fbshipit-source-id: 72df756a3a1f818688af66bbce4fc9fc02f1f505