[inductor] Implement magic methods on IR values (#101076)
This wraps `ops` into an `OpsWrapper` object which wraps any returned
IR values into an `OpsValue` instance. This allows magic methods to
be implemented and means lowerings can write mathematical expressions much more
fluently. So instead of
```python
ops.add(ops.mul(ops.mul(ops.sub(ops.mul(_Ap2, x), _Ap3), x), x), _1)
```
we can write
```python
(_Ap2 * x - _Ap3) * x * x + _1
```
And it will translate to the equivalent `ops` calls.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101076
Approved by: https://github.com/lezcano, https://github.com/ngimel
