### Summary

The mobile build from master was broken for two weeks due to a runtime error:

```shell
libc++abi.dylib: terminating with uncaught exception of type torch::jit::script::ErrorReport: 
Unknown builtin op: aten::_adaptive_avg_pool2d_backward.
Could not find any similar ops to aten::_adaptive_avg_pool2d_backward. This op may not exist or may not be currently supported in TorchScript.
:
at <string>:9:28
                grad_self = grad.expand(self.size()) / (self_size[-1] * self_size[-2])
            else:
                grad_self = torch._adaptive_avg_pool2d_backward(grad, self)
                            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE

            return grad_self
```
Since we've disabled the autograd for the opensourced version, the `backward` ops won't get registered by JIT. 

When `forward` runs, `ProfilingGraphExecutorImpl::getPlanFor(Stack& stack)` gets called, which will try to invoke `differentiate` according to the value of `needs_gradient`. 

```cpp
 // TODO: insert grad propagation
  bool needs_gradient = getProfilingMode()
      ? needsGradientInProfilingMode(copy->block())
      : true;
  if (needs_gradient) {
    // for Simple Executor skip creating autodiff graphs
    // and let autograd handle backward for us
    if (getProfilingMode()) {
      auto diff_nodes = CreateAutodiffSubgraphs(
        copy,
        getAutodiffSubgraphInlining() ? autodiffSubgraphNodeThreshold : 1);
      for (Node *dnode : diff_nodes) {
        auto diff_graph = std::move(dnode->g(attr::Subgraph));
        Gradient gradient = differentiate(diff_graph);
        runOptimization(gradient.f);
        // run non diff optimization on the forward graph
        runNondiffOptimization(gradient.f);
        packGradient(gradient, dnode);
      }
      InlineAutodiffSubgraphs(copy, getAutodiffSubgraphInlining()
                                        ? autodiffSubgraphInlineThreshold
                                        : 1);
    }
  } else {
    runNondiffOptimization(copy);
  }
```

The fix is sort of a workaround to get rid of the `profiling_mode`. Feel free to drop comments, if there is a better way to fix it.

### Test Plan

- The error above disappears
- Don't break CI

cc @AshkanAliabadi