[optim]: Adagrad with sparse grad is not on multi_tensor testing path + bug

### 🐛 Describe the bug

Bug: should not return here, just call the function
https://github.com/pytorch/pytorch/blob/b25e2f2b5e6bb95fc83a81bca23706bd6f08ae76/torch/optim/adagrad.py#L327

Solution: fix this, and add adagrad sparse to testing path (test_optim, inductor test)


### Versions

Main

cc @vincentqb @jbschlosser @albanD @janeyx99 @crcrpar