Orca datatype convert issue

[hkvision]

More details at this PR: #4444

The data type conversion from RDD to numpy is complete. But adding the missing data type conversions will cause failures/modifications in many unit tests.

The main issue is that:

• If user don't use numpy but simply python for their data; when converting to dataframe, the data types will always be double and long, no float/int.
• Users may not be aware of the behavior above and if model expects float/int, will throw error when they get double/long as input.
• For example in many of our unit tests, the dataframe type is wrong/not expected (actually array of double not float); and the data type conversion is also not complete (all convert to ndarray of float); but the two errors result the final input to be correct (torch tensor of float).