value_and_grad returns None gradients with thunder.jit

@kshitij12345 is already working on this.

Since Apr 24, 2024.

Note that it works with the deprecated `thunder.compile`.
```python
import torch
from thunder.core.transforms import value_and_grad
import thunder

def model(x, w1):
return x + w1

inp = torch.randn(1, 1) # doesn't matter if requires_grad is True/False
w1 = torch.randn(1, 1) # doesn't matter if requires_grad is True/False

print(thunder.compile(value_and_grad(model), disable_preprocessing=True)(inp, w1))
print(thunder.jit(value_and_grad(model))(inp, w1))
```

Output:
```
(tensor([[0.4349]]), (tensor([[1.]]), tensor([[1.]])))
(tensor([[0.4349]]), (None, None))
```