[BE] Migrate dtype_abbrs into one location (#152229)

Namely `torch.utils._dtype_abbrs.dtype_abbrs`

Before that it was defined in various forms of completeness in
c02edba863/torch/fx/graph.py (L215),
c02edba863/torch/testing/_internal/common_utils.py (L5226)
 and c02edba863/torch/testing/_internal/logging_tensor.py (L17)

TODO:
 - Add linter that `torch.testing._internal` module is not referenced from any of the public facing APIs, as it can have extra dependencies such as `expect_test`

Fixes https://github.com/pytorch/pytorch/issues/152225

Pull Request resolved: https://github.com/pytorch/pytorch/pull/152229
Approved by: https://github.com/clee2000, https://github.com/Skylion007

torch/testing/_internal/common_utils.py

@@ -5223,23 +5223,6 @@ def dtype_name(dtype):
     return str(dtype).split('.')[1]
 
 
-dtype_abbrs = {
-    torch.bfloat16: 'bf16',
-    torch.float64: 'f64',
-    torch.float32: 'f32',
-    torch.float16: 'f16',
-    torch.complex32: 'c32',
-    torch.complex64: 'c64',
-    torch.complex128: 'c128',
-    torch.int8: 'i8',
-    torch.int16: 'i16',
-    torch.int32: 'i32',
-    torch.int64: 'i64',
-    torch.bool: 'b8',
-    torch.uint8: 'u8',
-}
-
-
 @functools.lru_cache
 def get_cycles_per_ms() -> float:
     """Measure and return approximate number of cycles per millisecond for torch.cuda._sleep