Fix failing correctness and strings tests

Summary:
Fix failing tests for `TestConvertBoundingBoxFormat::test_correctness` and `TestConvertBoundingBoxFormat::test_strings`

Test Plan:
```bash
pytest test/test_transforms_v2.py -vvv -k "TestConvertBoundingBoxFormat"
...
87 passed, 32 skipped, 6964 deselected in 1.59s
```

Please note that the following tests `test/test_transforms_v2.py::TestConvertBoundingBoxFormat::test_correctness` were failing for previous format "CXCYWH", "XYXY" and "XYWH" for specific generated boxes.

```python
old_format = tv_tensors.BoundingBoxFormat.CXCYWH
new_format = tv_tensors.BoundingBoxFormat.XYXY

dtype = torch.int64
fn_type = "functional"
device = torch.device("cpu")


# bounding_boxes = make_bounding_boxes(format=old_format, dtype=dtype, device=device)
bounding_boxes = tv_tensors.BoundingBoxes([[ 5,  6, 10, 13]], format=tv_tensors.BoundingBoxFormat.CXCYWH, canvas_size=(17, 11))

if fn_type == "functional":
    fn = functools.partial(F.convert_bounding_box_format, new_format=new_format)
else:
    fn = transforms.ConvertBoundingBoxFormat(format=new_format)

actual = fn(bounding_boxes)
expected = _reference_convert_bounding_box_format(bounding_boxes, new_format)

assert_equal(actual, expected)
```

Author: AntoineSimoulin

test/common_utils.py

@@ -421,9 +421,9 @@ def sample_position(values, max_value):
     dtype = dtype or torch.float32
 
     h, w = [torch.randint(1, s, (num_boxes,)) for s in canvas_size]
-    r = -360 * torch.rand((num_boxes,)) + 180
     y = sample_position(h, canvas_size[0])
     x = sample_position(w, canvas_size[1])
+    r = -360 * torch.rand((num_boxes,)) + 180
 
     if format is tv_tensors.BoundingBoxFormat.XYWH:
         parts = (x, y, w, h)

torchvision/transforms/v2/functional/_meta.py

@@ -183,9 +183,9 @@ def _cxcywhr_to_xywhr(cxcywhr: torch.Tensor, inplace: bool) -> torch.Tensor:
     half_wh = cxcywhr[..., 2:-1].div(-2, rounding_mode=None if cxcywhr.is_floating_point() else "floor").abs_()
     r_rad = cxcywhr[..., 4].mul(torch.pi).div(180.0)
     # (cx - width / 2 * cos - height / 2 * sin) = x1
-    cxcywhr[..., 0].sub_(half_wh[..., 0].mul(r_rad.cos()).add(half_wh[..., 1].mul(r_rad.sin())).to(cxcywhr.dtype))
+    cxcywhr[..., 0].sub_(half_wh[..., 0].mul(r_rad.cos()).to(cxcywhr.dtype)).sub_(half_wh[..., 1].mul(r_rad.sin()).to(cxcywhr.dtype))
     # (cy + width / 2 * sin - height / 2 * cos) = y1
-    cxcywhr[..., 1].add_(half_wh[..., 0].mul(r_rad.sin()).sub(half_wh[..., 1].mul(r_rad.cos())).to(cxcywhr.dtype))
+    cxcywhr[..., 1].add_(half_wh[..., 0].mul(r_rad.sin()).to(cxcywhr.dtype)).sub_(half_wh[..., 1].mul(r_rad.cos()).to(cxcywhr.dtype))
 
     return cxcywhr
 
@@ -197,9 +197,9 @@ def _xywhr_to_cxcywhr(xywhr: torch.Tensor, inplace: bool) -> torch.Tensor:
     half_wh = xywhr[..., 2:-1].div(-2, rounding_mode=None if xywhr.is_floating_point() else "floor").abs_()
     r_rad = xywhr[..., 4].mul(torch.pi).div(180.0)
     # (x1 + width / 2 * cos + height / 2 * sin) = cx
-    xywhr[..., 0].add_(half_wh[..., 0].mul(r_rad.cos()).add(half_wh[..., 1].mul(r_rad.sin())).to(xywhr.dtype))
+    xywhr[..., 0].add_(half_wh[..., 0].mul(r_rad.cos()).to(xywhr.dtype)).add_(half_wh[..., 1].mul(r_rad.sin()).to(xywhr.dtype))
     # (y1 - width / 2 * sin + height / 2 * cos) = cy 
-    xywhr[..., 1].add_(half_wh[..., 1].mul(r_rad.cos()).sub(half_wh[..., 0].mul(r_rad.sin())).to(xywhr.dtype))
+    xywhr[..., 1].sub_(half_wh[..., 0].mul(r_rad.sin()).to(xywhr.dtype)).add_(half_wh[..., 1].mul(r_rad.cos()).to(xywhr.dtype))
 
     return xywhr