Rewrite lerp operator to use TensorIterator and support compile-time vectorization. (#22038)
Summary:
Get benefit from the compile time vectorization and multi-threading.
Before:
```python
In [1]: import torch
In [2]: x = torch.randn(1000000)
In [3]: y = torch.randn(1000000)
In [4]: w = 0.7
In [5]: timeit torch.lerp(x, y, w)
2.29 ms ± 23.9 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
```
After:
```python
In [1]: import torch
In [2]: x = torch.randn(1000000)
In [3]: y = torch.randn(1000000)
In [4]: w = 0.7
In [5]: timeit torch.lerp(x, y, w)
452 µs ± 1.81 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
```
After with multi-processing:
```python
In [1]: import torch
In [2]: x = torch.randn(1000000)
In [3]: y = torch.randn(1000000)
In [4]: w = 0.7
In [5]: timeit torch.lerp(x, y, w)
167 µs ± 48.8 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22038
Differential Revision: D15941468
Pulled By: VitalyFedyunin
fbshipit-source-id: fa8a5126187df4e6c849452e035b00b22be25739
