[MPS] Unregister put_() op due to lack of implementation (pytorch#94231)

Currently, the `put_()` is not implemented on MPS backend, so this patch will unregister it and insert it into blocklist of TestConsistency.
Pull Request resolved: pytorch#94231
Approved by: https://github.com/kulinseth

aten/src/ATen/native/native_functions.yaml

@@ -7434,7 +7434,7 @@
 - func: put_(Tensor(a!) self, Tensor index, Tensor source, bool accumulate=False) -> Tensor(a!)
   variants: method
   dispatch:
-    CPU, CUDA, MPS: put_
+    CPU, CUDA: put_
   autogen: put.out
 
 - func: put(Tensor self, Tensor index, Tensor source, bool accumulate=False) -> Tensor

test/test_mps.py

@@ -8701,6 +8701,9 @@ class TestConsistency(TestCase):
         # count_nonzero returns wrong results for these dtypes
         'nonzero': [torch.uint8, torch.float16],
 
+        # failures due to lack of op implementation on MPS backend
+        'put': ['torch.bool', 'torch.float16', 'torch.float32', 'torch.int16', 'torch.int32', 'torch.int64', 'torch.uint8'],
+
         # These were moved from ALLOWLIST to BLOCK as they are not working
         # locally
         'tile': ['torch.float16', 'torch.float32', 'torch.int16', 'torch.int32', 'torch.int64', 'torch.uint8'],