Make TensorCoreTiledLayout import more robust

Summary: Fixes pytorch#1908.
Previous attempts (pytorch#1886)
to fix this issue still break in local settings, so it's more
robust and simpler to just try catch the import error.

Test Plan:
```
from torchtune.training.quantization import *
```

Author: andrewor14

torchtune/training/quantization.py

@@ -6,11 +6,11 @@
 
 from typing import Callable, Optional
 
-from torchtune.utils._import_guard import _USE_NEW_TENSOR_CORE_TILED_LAYOUT_API
-
-if _USE_NEW_TENSOR_CORE_TILED_LAYOUT_API:
+try:
+    # torchao 0.7+
     from torchao.dtypes import TensorCoreTiledLayout
-else:
+except ImportError:
+    # torchao 0.6 and before
     from torchao.dtypes import TensorCoreTiledLayoutType as TensorCoreTiledLayout
 
 from torchao.quantization import (

torchtune/utils/_import_guard.py

@@ -6,7 +6,7 @@
 
 import torch
 import torchao
-from torchtune.utils._version import _is_fbcode, _nightly_version_ge, torch_version_ge
+from torchtune.utils._version import torch_version_ge
 
 # We can only use flex attention / BlockMask if torch version >= 2.5.0 and GPU is Turing / SM75 and above
 _SUPPORTS_FLEX_ATTENTION = (
@@ -16,14 +16,3 @@
 )
 
 torchao_version = torchao.__version__
-
-_USE_NEW_TENSOR_CORE_TILED_LAYOUT_API = _is_fbcode() or (
-    not _is_fbcode()
-    and (
-        ("dev" not in torchao_version and torchao_version >= "0.7.0")
-        or (
-            "dev" in torchao_version
-            and _nightly_version_ge(torchao_version, "2024-10-10")
-        )
-    )
-)