[BE] [MPS] Fix `out` resize logic in `torch.where` (#121476)

By deleting `where_mps`  and registering MPS dispatch for `where_kernel`.
As result of this change resizing and type-checking logic is shared between MPS, CPU and  CUDA backends.

Add test_case to `TestMPS.test_where` (that should eventually be removed, when `out` OpInfo testing is enabled for MPS
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121476
Approved by: https://github.com/albanD, https://github.com/Skylion007
ghstack dependencies: #121473, #121494

test/test_mps.py

diff --git a/test/test_mps.py b/test/test_mps.py
index af26cbb..3c5e47c 100644
--- a/test/test_mps.py
+++ b/test/test_mps.py
@@ -7475,6 +7475,15 @@
         helper((2, 3), (5, 2, 3), (2, 3))
         helper((2, 3), (2, 3), (5, 2, 3))
         helper((2, 3), (5, 2, 3), (6, 5, 2, 3))
+        # Test that output is correctly resizes
+        # TODO: Remove me when out OpInfo testing is enabled on MPS
+        output = torch.tensor(0.0, device="mps")
+        cond = torch.randint(2, (3, 3), dtype=torch.bool, device="mps")
+        inp = torch.rand(3, 3, device="mps")
+        other = torch.rand(3, 3, device="mps")
+        out = torch.where(cond, inp, other, out=output)
+        self.assertEqual(id(out), id(output))
+        self.assertEqual(out.shape, (3, 3))
 
     # Test normal
     def test_normal(self):