dtype inconsistencies when dividing/rounding tensors

Nobody has claimed this yet.

## 🐛 Bug

This bug was found when submitting a PR for `torch.div`(https://github.com/Lightning-AI/lightning-thunder/pull/441). Specifically, there was a dtype mismatch when rounding and dividing tensors.

### To Reproduce

The torch.div opinfo currently xfails the `test_core_vs_torch_consistency_div` test with nvfuser. However, by removing the xfail with nvfuser in div_opinfo,, it will throw the following error message:

```
FAILED thunder/tests/test_ops.py::test_core_vs_torch_consistency_div_nvfuser_cuda_int32 - AssertionError: The values for attribute 'dtype' do not match: torch.float32 != torch.int32.
FAILED thunder/tests/test_ops.py::test_core_vs_torch_consistency_div_nvfuser_cuda_int64 - AssertionError: The values for attribute 'dtype' do not match: torch.float32 != torch.int64.
FAILED thunder/tests/test_ops.py::test_core_vs_torch_consistency_div_nvfuser_cuda_float32 - AssertionError: The values for attribute 'dtype' do not match: torch.float64 != torch.float32.
```

### Expected behavior

It is expected to for the tensors/numbers to have the same dtype when using nvFuser.

### Additional context

This can potentially be addressed by adding an OpInfo for `prims.div`.

cc. @mruberry

Start with the torch.div OpInfo and the xfailed test_core_vs_torch_consistency_div tests in thunder/tests/test_ops.py, focusing on the nvFuser CUDA cases listed in the report. Check whether adding an OpInfo for prims.div addresses the rounding and division dtype mismatches. Done means the affected tests no longer report differing dtypes for the compared tensors.

Written by the indexing model from the issue text.