[ROCm] Add torch/_rocm_init.py to .gitignore. (#159806)

Follow-up to https://github.com/pytorch/pytorch/pull/155285.

Build scripts like https://github.com/ROCm/TheRock/blob/main/external-builds/pytorch/build_prod_wheels.py generate this file with contents like:

```python
def initialize():
    import rocm_sdk
    rocm_sdk.initialize_process(
        preload_shortnames=['amd_comgr', 'amdhip64', 'hiprtc', 'hipblas', 'hipfft', 'hiprand', 'hipsparse', 'hipsolver', 'hipblaslt', 'miopen'],
        check_version='7.0.0rc20250804')
```

We may also have https://github.com/pytorch/pytorch/blob/main/tools/amd_build/build_amd.py do the same thing as more of that build support moves here into the upstream PyTorch repository itself (see https://github.com/pytorch/pytorch/issues/159520).

This file is then loaded if present here: a7f3bdf550/torch/__init__.py (L145-L157)

Given that the file is generated by build scripts, I think adding it to `.gitignore` makes sense, as that will prevent accidental check-ins and keep local history cleaner.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/159806
Approved by: https://github.com/jeffdaily

.gitignore

@@ -146,6 +146,9 @@ merge_record.json
 torchgen/packaged/*
 !torchgen/packaged/README.md
 
+# This file is injected by ROCm build scripts to bootstrap in torch/__init__.py.
+torch/_rocm_init.py
+
 # IPython notebook checkpoints
 .ipynb_checkpoints