[LTC] Add nll_loss_forward TorchScript lowering (#63800)
Summary:
This patch added nll_loss_forward to TorchScript lowering. To be noted, since ATen treats
nll_loss as a hacky version of nll_loss_forward, here we change LazyTensor's NllLoss IR
to NllLossForward to follow the PyTorch conversion. And thus it diverges from XLA.
Test Plan:
[.../pytorch/lazy_tensor_core] test/cpp/build/test_ptltc --gtest_filter=AtenLtcTsTensorTest.TestNllLoss
[.../pytorch/lazy_tensor_core] test/cpp/build/test_ptltc --gtest_filter=AtenLtcTsTensorTest.TestNllLossBackward
