Aten scatter converter

[apbose]

Dependency of PR- #2519

[apbose]

The harness.py file is changed in this for the cases in which the index is passed in the forward function. torch scatter function takes only int64 inputs for index Couple of points

1. When passed as a variable outside the forward function, the get_trt_tensor function will cast to int64
2. When int64 is passed in the forward function, it throws error in the unified_dtype_converter in fx/utils.py. This is in TRTInterpretor.py while adding the placeholder nodes - self.ctx.net.add_input
3. Passing truncate_long_and_double in compilationsettings does not help since this would repair inputs in _compiler.py outside the TRTInterpreter and it would not help in harness.py
4. Had the tests been in the form of lowering tests, the line - https://github.com/pytorch/TensorRT/blob/16c031349c6a1af5a8408a817f2ef8542aa6f176/py/torch_tensorrt/dynamo/_compiler.py#L394 would create issue