Add ReplaceNaN benchmark as baseline (#46685)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/46685
as title
Test Plan:
caffe2
```
./buck-out/gen/caffe2/benchmarks/operator_benchmark/c2/replace_nan_test.par
# ----------------------------------------
# PyTorch/Caffe2 Operator Micro-benchmarks
# ----------------------------------------
# Tag : short
# Benchmarking Caffe2: replace_nan
WARNING: Logging before InitGoogleLogging() is written to STDERR
W1022 10:09:48.508246 1887813 init.h:137] Caffe2 GlobalInit should be run before any other API calls.
# Name: replace_nan_M16_N16_dtypefloat
# Input: M: 16, N: 16, dtype: float
Forward Execution Time (us) : 30.742
# Benchmarking Caffe2: replace_nan
# Name: replace_nan_M16_N16_dtypedouble
# Input: M: 16, N: 16, dtype: double
Forward Execution Time (us) : 29.135
# Benchmarking Caffe2: replace_nan
# Name: replace_nan_M64_N64_dtypefloat
# Input: M: 64, N: 64, dtype: float
Forward Execution Time (us) : 94.059
# Benchmarking Caffe2: replace_nan
# Name: replace_nan_M64_N64_dtypedouble
# Input: M: 64, N: 64, dtype: double
Forward Execution Time (us) : 93.569
```
Reviewed By: qizzzh, houseroad
Differential Revision: D24448483
fbshipit-source-id: 51574ca0eca6dba5828dfdc754193dba5a62954f
