New ruleset: torch load models with safetensor

Code similar to:

```
        $VAR_TOKENIZER = $TOKENIZER.from_pretrained("...")
```

Should be marked, the advised fix is the same code plus the `use_safetensors=True` helper

```
        $VAR_TOKENIZER = $TOKENIZER.from_pretrained("...", use_safetensors=True)
```
        