Fix correctness issues with Upsample 1D and 2D (#183)

* Fix correctness issues with Upsample 1D and 2D
- Implemented following new ops:
upsample_nearest1d_backward
upsample_nearest_exact1d
upsample_nearest_exact1d_backward
- Moved Upsample code from Shape.mm to Upsample.mm

* Corrections on some comments

* Fix the leak in cat_out_mps and clean up

Author: razarmehr

aten/src/ATen/mps/MPSFallback.mm

@@ -62,6 +62,7 @@ Tensor slow_conv2d_forward_mps(
   m.impl("linalg_vector_norm", torch::CppFunction::makeFromBoxedFunction<&mps_fallback>());
   m.impl("sgn.out", torch::CppFunction::makeFromBoxedFunction<&mps_fallback>());
   m.impl("_slow_conv2d_forward", slow_conv2d_forward_mps);
+  m.impl("upsample_nearest3d.vec", torch::CppFunction::makeFromBoxedFunction<&mps_fallback>());
 }
 
 } // namespace at