avoid torch.ops.load_library() calls

Summary:
Due to different buck root, code such as
```
torch.ops.load_library("//executorch/kernels/quantized:custom_ops_generated_lib")
```
is very hard to work both internally and in oss. Here I'm trying to leverage the `preload_deps` attribute in python_binary and python_unittest, and remove `torch.ops.load_library` calls. This should unblock these tests/binaries to be running in OSS.

Reviewed By: cccclai

Differential Revision: D48018133

fbshipit-source-id: 06dcf01a611e5b8877a71ebded5f9753db22a55f