Fix the missing device in _memory_profiler (#119751)
Fixes #119722,
1, added the missing device in
```
max_memory_allocated = torch.cuda.max_memory_allocated()
max_memory_reserved = torch.cuda.max_memory_reserved()
```
2, fix the device parameter to device_str. Based on [lines](https://github.com/pytorch/pytorch/blob/2bda6b4cb80ea46c09ef8123dd0255a26e7d157f/torch/profiler/profiler.py#L291), the input device are a string (device_str) for
```
self.mem_tl.export_memory_timeline_html
self.mem_tl.export_memory_timeline_raw
self.mem_tl.export_memory_timeline
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119751
Approved by: https://github.com/aaronenyeshi
