Add ONNX models

These are not perfect conversions. The Lite model differs from the PyTorch version by ~3%. The main model differs by ~1%.

4x-UltraSharpV2_Lite_fp16_op17.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b368dd0460421c3b3484a9a6855c07670f853abde3e0e5a6bfb72f2d5f8d9c50
+size 15281610

4x-UltraSharpV2_Lite_fp32_op17.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba692ad6c7b59bdebbaa9951c9ef5295a6d69e7444f1c46824c3cafdaab067a8
+size 29911078

4x-UltraSharpV2_fp16_op17.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3002bbbc4779a4f44db72a0bff0b6ace7ab65b7ec3431866c6c99b9de3a81d3
+size 31854429

4x-UltraSharpV2_fp32_op17.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c0201e3403745f39a9aa5273c50ed084cf6c4af5a71a4654c30252cf19bf0a5
+size 51800517