xla

984aa8d8 - [Functionalization] Undo lowering _convolution/convolution_backward

[Functionalization] Undo lowering _convolution/convolution_backward Summary: I somehow wrongly lowered _convolution/convolution_backward in the early stage. It then makes conv.backward disappear from the graph. Therefore, undoing that change and adds a test case for it. Test Plan: PJRT_DEVICE=TPU python test/test_operations.py -v -k test_conv2d_backward