fix: set codegen in _SplitterBase partitioner (#120361)
For graphs with single output, the expectation of torch.export / torch.compile graph_module output type is a single torch.tensor instead of a tuple.
However, after using `_SplitterBase` partitioner on these graph_module (obtained from torch.export/torch.compile), the resultant graph module will return a tuple of tensors, in this case `(output,)`.
This PR adds codegen to the graphs produced by `_SplitterBase` partitioner. Setting this will ensure pytree unflatten nodes will be added automatically to handle unflattening of the output to return single outputs directly.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120361
Approved by: https://github.com/angelayi
