call _upcast to consider overflow for computing giou loss (#5685)

* call _upcast to consider overflow

giou loss is weak at overflow problem because it computes area of box.

* cast datatype to float

* add ":" to if

* lint Test for membership should be 'not in' (E713)

torchvision/ops/giou_loss.py

 import torch
+from torch import Tensor
+
+
+def _upcast(t: Tensor) -> Tensor:
+    # Protects from numerical overflows in multiplications by upcasting to the equivalent higher type
+    if t.dtype not in (torch.float32, torch.float64):
+        return t.float()
+    return t
 
 
 def generalized_box_iou_loss(
@@ -34,6 +42,8 @@ def generalized_box_iou_loss(
         https://arxiv.org/abs/1902.09630
     """
 
+    boxes1 = _upcast(boxes1)
+    boxes2 = _upcast(boxes2)
     x1, y1, x2, y2 = boxes1.unbind(dim=-1)
     x1g, y1g, x2g, y2g = boxes2.unbind(dim=-1)